diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/Kconfig | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device.c | 123 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 11 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_events.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 5 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 8 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 45 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 6 |
12 files changed, 141 insertions, 67 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/Kconfig b/drivers/gpu/drm/amd/amdkfd/Kconfig index 62e88e5362e9..16e12c9913f9 100644 --- a/drivers/gpu/drm/amd/amdkfd/Kconfig +++ b/drivers/gpu/drm/amd/amdkfd/Kconfig @@ -5,7 +5,7 @@ config HSA_AMD bool "HSA kernel driver for AMD GPU devices" - depends on DRM_AMDGPU && (X86_64 || ARM64 || PPC64 || (RISCV && 64BIT)) + depends on DRM_AMDGPU && (X86_64 || ARM64 || PPC64 || (RISCV && 64BIT) || (LOONGARCH && 64BIT)) select HMM_MIRROR select MMU_NOTIFIER select DRM_AMDGPU_USERPTR diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index a2149afa5803..828a9ceef1e7 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -22,7 +22,6 @@ */ #include <linux/device.h> -#include <linux/export.h> #include <linux/err.h> #include <linux/fs.h> #include <linux/file.h> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index bf0854bd5555..7e749f9b6d69 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -971,7 +971,7 @@ int kgd2kfd_pre_reset(struct kfd_dev *kfd, kfd_smi_event_update_gpu_reset(node, false, reset_context); } - kgd2kfd_suspend(kfd, false); + kgd2kfd_suspend(kfd, true); for (i = 0; i < kfd->num_nodes; i++) kfd_signal_reset_event(kfd->nodes[i]); @@ -1013,13 +1013,33 @@ int kgd2kfd_post_reset(struct kfd_dev *kfd) return 0; } -bool kfd_is_locked(void) +bool kfd_is_locked(struct kfd_dev *kfd) { + uint8_t id = 0; + struct kfd_node *dev; + lockdep_assert_held(&kfd_processes_mutex); - return (kfd_locked > 0); + + /* check reset/suspend lock */ + if (kfd_locked > 0) + return true; + + if (kfd) + return kfd->kfd_dev_lock > 0; + + /* check lock on all cgroup accessible devices */ + while (kfd_topology_enum_kfd_devices(id++, &dev) == 0) { + if (!dev || kfd_devcgroup_check_permission(dev)) + continue; + + if (dev->kfd->kfd_dev_lock > 0) + return true; + } + + return false; } -void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm) +void kgd2kfd_suspend(struct kfd_dev *kfd, bool suspend_proc) { struct kfd_node *node; int i; @@ -1027,14 +1047,8 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm) if (!kfd->init_complete) return; - /* for runtime suspend, skip locking kfd */ - if (!run_pm) { - mutex_lock(&kfd_processes_mutex); - /* For first KFD device suspend all the KFD processes */ - if (++kfd_locked == 1) - kfd_suspend_all_processes(); - mutex_unlock(&kfd_processes_mutex); - } + if (suspend_proc) + kgd2kfd_suspend_process(kfd); for (i = 0; i < kfd->num_nodes; i++) { node = kfd->nodes[i]; @@ -1042,7 +1056,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm) } } -int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm) +int kgd2kfd_resume(struct kfd_dev *kfd, bool resume_proc) { int ret, i; @@ -1055,14 +1069,36 @@ int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm) return ret; } - /* for runtime resume, skip unlocking kfd */ - if (!run_pm) { - mutex_lock(&kfd_processes_mutex); - if (--kfd_locked == 0) - ret = kfd_resume_all_processes(); - WARN_ONCE(kfd_locked < 0, "KFD suspend / resume ref. error"); - mutex_unlock(&kfd_processes_mutex); - } + if (resume_proc) + ret = kgd2kfd_resume_process(kfd); + + return ret; +} + +void kgd2kfd_suspend_process(struct kfd_dev *kfd) +{ + if (!kfd->init_complete) + return; + + mutex_lock(&kfd_processes_mutex); + /* For first KFD device suspend all the KFD processes */ + if (++kfd_locked == 1) + kfd_suspend_all_processes(); + mutex_unlock(&kfd_processes_mutex); +} + +int kgd2kfd_resume_process(struct kfd_dev *kfd) +{ + int ret = 0; + + if (!kfd->init_complete) + return 0; + + mutex_lock(&kfd_processes_mutex); + if (--kfd_locked == 0) + ret = kfd_resume_all_processes(); + WARN_ONCE(kfd_locked < 0, "KFD suspend / resume ref. error"); + mutex_unlock(&kfd_processes_mutex); return ret; } @@ -1442,24 +1478,53 @@ unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_node *node) kfd_get_num_sdma_engines(node); } -int kgd2kfd_check_and_lock_kfd(void) +int kgd2kfd_check_and_lock_kfd(struct kfd_dev *kfd) { + struct kfd_process *p; + int r = 0, temp, idx; + mutex_lock(&kfd_processes_mutex); - if (!hash_empty(kfd_processes_table) || kfd_is_locked()) { - mutex_unlock(&kfd_processes_mutex); - return -EBUSY; + + if (hash_empty(kfd_processes_table) && !kfd_is_locked(kfd)) + goto out; + + /* fail under system reset/resume or kfd device is partition switching. */ + if (kfd_is_locked(kfd)) { + r = -EBUSY; + goto out; + } + + /* + * ensure all running processes are cgroup excluded from device before mode switch. + * i.e. no pdd was created on the process socket. + */ + idx = srcu_read_lock(&kfd_processes_srcu); + hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) { + int i; + + for (i = 0; i < p->n_pdds; i++) { + if (p->pdds[i]->dev->kfd != kfd) + continue; + + r = -EBUSY; + goto proc_check_unlock; + } } - ++kfd_locked; +proc_check_unlock: + srcu_read_unlock(&kfd_processes_srcu, idx); +out: + if (!r) + ++kfd->kfd_dev_lock; mutex_unlock(&kfd_processes_mutex); - return 0; + return r; } -void kgd2kfd_unlock_kfd(void) +void kgd2kfd_unlock_kfd(struct kfd_dev *kfd) { mutex_lock(&kfd_processes_mutex); - --kfd_locked; + --kfd->kfd_dev_lock; mutex_unlock(&kfd_processes_mutex); } 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 76359c6a3f3a..2d91027e2a74 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -2312,7 +2312,7 @@ static int reset_hung_queues_sdma(struct device_queue_manager *dqm) continue; /* Reset engine and check. */ - if (amdgpu_sdma_reset_engine(dqm->dev->adev, i) || + if (amdgpu_sdma_reset_engine(dqm->dev->adev, i, false) || dqm->dev->kfd2kgd->hqd_sdma_get_doorbell(dqm->dev->adev, i, j) || !set_sdma_queue_as_reset(dqm, doorbell_off)) { r = -ENOTRECOVERABLE; @@ -2339,9 +2339,18 @@ reset_fail: static int reset_queues_on_hws_hang(struct device_queue_manager *dqm, bool is_sdma) { + struct amdgpu_device *adev = dqm->dev->adev; + while (halt_if_hws_hang) schedule(); + if (adev->debug_disable_gpu_ring_reset) { + dev_info_once(adev->dev, + "%s queue hung, but ring reset disabled", + is_sdma ? "sdma" : "compute"); + + return -EPERM; + } if (!amdgpu_gpu_recovery) return -ENOTRECOVERABLE; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c index 2b294ada3ec0..82905f3e54dd 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -1302,7 +1302,7 @@ void kfd_signal_reset_event(struct kfd_node *dev) if (ti) { dev_err(dev->adev->dev, "Queues reset on process %s tid %d thread %s pid %d\n", - ti->process_name, ti->tgid, ti->task_name, ti->pid); + ti->process_name, ti->tgid, ti->task.comm, ti->task.pid); amdgpu_vm_put_task_info(ti); } } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c index dbcb60eb54b2..1d170dc50df3 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c @@ -23,7 +23,6 @@ */ #include <linux/device.h> -#include <linux/export.h> #include <linux/err.h> #include <linux/fs.h> #include <linux/sched.h> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c index 8fa6489b6f5d..505036968a77 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c @@ -240,7 +240,7 @@ static int pm_map_queues_v9(struct packet_manager *pm, uint32_t *buffer, packet->bitfields2.engine_sel = engine_sel__mes_map_queues__compute_vi; - packet->bitfields2.gws_control_queue = q->gws ? 1 : 0; + packet->bitfields2.gws_control_queue = q->properties.is_gws ? 1 : 0; packet->bitfields2.extended_engine_sel = extended_engine_sel__mes_map_queues__legacy_engine_sel; packet->bitfields2.queue_type = diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index d221c58dccc3..67694bcd9464 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -372,6 +372,9 @@ struct kfd_dev { /* bitmap for dynamic doorbell allocation from doorbell object */ unsigned long *doorbell_bitmap; + + /* for dynamic partitioning */ + int kfd_dev_lock; }; enum kfd_mempool { @@ -1536,7 +1539,7 @@ static inline bool kfd_flush_tlb_after_unmap(struct kfd_dev *dev) int kfd_send_exception_to_runtime(struct kfd_process *p, unsigned int queue_id, uint64_t error_reason); -bool kfd_is_locked(void); +bool kfd_is_locked(struct kfd_dev *kfd); /* Compute profile */ void kfd_inc_compute_active(struct kfd_node *dev); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 722ac1662bdc..5be28c6c4f6a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -854,7 +854,7 @@ struct kfd_process *kfd_create_process(struct task_struct *thread) */ mutex_lock(&kfd_processes_mutex); - if (kfd_is_locked()) { + if (kfd_is_locked(NULL)) { pr_debug("KFD is locked! Cannot create process"); process = ERR_PTR(-EINVAL); goto out; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c index 83d9384ac815..a499449fcb06 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c @@ -253,9 +253,9 @@ void kfd_smi_event_update_vmfault(struct kfd_node *dev, uint16_t pasid) task_info = amdgpu_vm_get_task_info_pasid(dev->adev, pasid); if (task_info) { /* Report VM faults from user applications, not retry from kernel */ - if (task_info->pid) + if (task_info->task.pid) kfd_smi_event_add(0, dev, KFD_SMI_EVENT_VMFAULT, KFD_EVENT_FMT_VMFAULT( - task_info->pid, task_info->task_name)); + task_info->task.pid, task_info->task.comm)); amdgpu_vm_put_task_info(task_info); } } @@ -359,8 +359,8 @@ void kfd_smi_event_process(struct kfd_process_device *pdd, bool start) kfd_smi_event_add(0, pdd->dev, start ? KFD_SMI_EVENT_PROCESS_START : KFD_SMI_EVENT_PROCESS_END, - KFD_EVENT_FMT_PROCESS(task_info->pid, - task_info->task_name)); + KFD_EVENT_FMT_PROCESS(task_info->task.pid, + task_info->task.comm)); amdgpu_vm_put_task_info(task_info); } } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 865dca2547de..a0f22ea6d15a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1171,13 +1171,12 @@ svm_range_split_head(struct svm_range *prange, uint64_t new_start, } static void -svm_range_add_child(struct svm_range *prange, struct mm_struct *mm, - struct svm_range *pchild, enum svm_work_list_ops op) +svm_range_add_child(struct svm_range *prange, struct svm_range *pchild, enum svm_work_list_ops op) { pr_debug("add child 0x%p [0x%lx 0x%lx] to prange 0x%p child list %d\n", pchild, pchild->start, pchild->last, prange, op); - pchild->work_item.mm = mm; + pchild->work_item.mm = NULL; pchild->work_item.op = op; list_add_tail(&pchild->child_list, &prange->child_list); } @@ -1278,7 +1277,7 @@ svm_range_get_pte_flags(struct kfd_node *node, mapping_flags |= ext_coherent ? AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC; /* system memory accessed by the dGPU */ } else { - if (gc_ip_version < IP_VERSION(9, 5, 0)) + if (gc_ip_version < IP_VERSION(9, 5, 0) || ext_coherent) mapping_flags |= AMDGPU_VM_MTYPE_UC; else mapping_flags |= AMDGPU_VM_MTYPE_NC; @@ -2394,15 +2393,17 @@ svm_range_add_list_work(struct svm_range_list *svms, struct svm_range *prange, prange->work_item.op != SVM_OP_UNMAP_RANGE) prange->work_item.op = op; } else { - prange->work_item.op = op; - - /* Pairs with mmput in deferred_list_work */ - mmget(mm); - prange->work_item.mm = mm; - list_add_tail(&prange->deferred_list, - &prange->svms->deferred_range_list); - pr_debug("add prange 0x%p [0x%lx 0x%lx] to work list op %d\n", - prange, prange->start, prange->last, op); + /* Pairs with mmput in deferred_list_work. + * If process is exiting and mm is gone, don't update mmu notifier. + */ + if (mmget_not_zero(mm)) { + prange->work_item.mm = mm; + prange->work_item.op = op; + list_add_tail(&prange->deferred_list, + &prange->svms->deferred_range_list); + pr_debug("add prange 0x%p [0x%lx 0x%lx] to work list op %d\n", + prange, prange->start, prange->last, op); + } } spin_unlock(&svms->deferred_list_lock); } @@ -2416,8 +2417,7 @@ void schedule_deferred_list_work(struct svm_range_list *svms) } static void -svm_range_unmap_split(struct mm_struct *mm, struct svm_range *parent, - struct svm_range *prange, unsigned long start, +svm_range_unmap_split(struct svm_range *parent, struct svm_range *prange, unsigned long start, unsigned long last) { struct svm_range *head; @@ -2438,12 +2438,12 @@ svm_range_unmap_split(struct mm_struct *mm, struct svm_range *parent, svm_range_split(tail, last + 1, tail->last, &head); if (head != prange && tail != prange) { - svm_range_add_child(parent, mm, head, SVM_OP_UNMAP_RANGE); - svm_range_add_child(parent, mm, tail, SVM_OP_ADD_RANGE); + svm_range_add_child(parent, head, SVM_OP_UNMAP_RANGE); + svm_range_add_child(parent, tail, SVM_OP_ADD_RANGE); } else if (tail != prange) { - svm_range_add_child(parent, mm, tail, SVM_OP_UNMAP_RANGE); + svm_range_add_child(parent, tail, SVM_OP_UNMAP_RANGE); } else if (head != prange) { - svm_range_add_child(parent, mm, head, SVM_OP_UNMAP_RANGE); + svm_range_add_child(parent, head, SVM_OP_UNMAP_RANGE); } else if (parent != prange) { prange->work_item.op = SVM_OP_UNMAP_RANGE; } @@ -2520,14 +2520,14 @@ svm_range_unmap_from_cpu(struct mm_struct *mm, struct svm_range *prange, l = min(last, pchild->last); if (l >= s) svm_range_unmap_from_gpus(pchild, s, l, trigger); - svm_range_unmap_split(mm, prange, pchild, start, last); + svm_range_unmap_split(prange, pchild, start, last); mutex_unlock(&pchild->lock); } s = max(start, prange->start); l = min(last, prange->last); if (l >= s) svm_range_unmap_from_gpus(prange, s, l, trigger); - svm_range_unmap_split(mm, prange, prange, start, last); + svm_range_unmap_split(prange, prange, start, last); if (unmap_parent) svm_range_add_list_work(svms, prange, mm, SVM_OP_UNMAP_RANGE); @@ -2570,8 +2570,6 @@ svm_range_cpu_invalidate_pagetables(struct mmu_interval_notifier *mni, if (range->event == MMU_NOTIFY_RELEASE) return true; - if (!mmget_not_zero(mni->mm)) - return true; start = mni->interval_tree.start; last = mni->interval_tree.last; @@ -2598,7 +2596,6 @@ svm_range_cpu_invalidate_pagetables(struct mmu_interval_notifier *mni, } svm_range_unlock(prange); - mmput(mni->mm); return true; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index baa2374acdeb..4ec73f33535e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -510,6 +510,10 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, dev->node_props.capability |= HSA_CAP_AQL_QUEUE_DOUBLE_MAP; + if (KFD_GC_VERSION(dev->gpu) < IP_VERSION(10, 0, 0) && + (dev->gpu->adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE)) + dev->node_props.capability2 |= HSA_CAP2_PER_SDMA_QUEUE_RESET_SUPPORTED; + sysfs_show_32bit_prop(buffer, offs, "max_engine_clk_fcompute", dev->node_props.max_engine_clk_fcompute); @@ -2008,8 +2012,6 @@ static void kfd_topology_set_capabilities(struct kfd_topology_device *dev) if (!amdgpu_sriov_vf(dev->gpu->adev)) dev->node_props.capability |= HSA_CAP_PER_QUEUE_RESET_SUPPORTED; - if (dev->gpu->adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE) - dev->node_props.capability2 |= HSA_CAP2_PER_SDMA_QUEUE_RESET_SUPPORTED; } else { dev->node_props.debug_prop |= HSA_DBG_WATCH_ADDR_MASK_LO_BIT_GFX10 | HSA_DBG_WATCH_ADDR_MASK_HI_BIT; |