diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 12 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 85 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_events.c | 11 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 7 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.h | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 12 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_queue.c | 12 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 22 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 4 |
11 files changed, 75 insertions, 93 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 0f0719528bcc..22925df6a791 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -2826,7 +2826,7 @@ retry: static int runtime_disable(struct kfd_process *p) { - int i = 0, ret; + int i = 0, ret = 0; bool was_enabled = p->runtime_info.runtime_state == DEBUG_RUNTIME_STATE_ENABLED; p->runtime_info.runtime_state = DEBUG_RUNTIME_STATE_DISABLED; @@ -2863,6 +2863,7 @@ static int runtime_disable(struct kfd_process *p) /* disable ttmp setup */ for (i = 0; i < p->n_pdds; i++) { struct kfd_process_device *pdd = p->pdds[i]; + int last_err = 0; if (kfd_dbg_is_per_vmid_supported(pdd->dev)) { pdd->spi_dbg_override = @@ -2872,14 +2873,17 @@ static int runtime_disable(struct kfd_process *p) pdd->dev->vm_info.last_vmid_kfd); if (!pdd->dev->kfd->shared_resources.enable_mes) - debug_refresh_runlist(pdd->dev->dqm); + last_err = debug_refresh_runlist(pdd->dev->dqm); else - kfd_dbg_set_mes_debug_mode(pdd, + last_err = kfd_dbg_set_mes_debug_mode(pdd, !kfd_dbg_has_cwsr_workaround(pdd->dev)); + + if (last_err) + ret = last_err; } } - return 0; + return ret; } static int kfd_ioctl_runtime_enable(struct file *filep, struct kfd_process *p, void *data) 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 6c5c7c1bf5ed..d7a2e7178ea9 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -1209,6 +1209,15 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, pr_debug_ratelimited("Evicting process pid %d queues\n", pdd->process->lead_thread->pid); + if (dqm->dev->kfd->shared_resources.enable_mes) { + pdd->last_evict_timestamp = get_jiffies_64(); + retval = suspend_all_queues_mes(dqm); + if (retval) { + dev_err(dev, "Suspending all queues failed"); + goto out; + } + } + /* Mark all queues as evicted. Deactivate all active queues on * the qpd. */ @@ -1221,23 +1230,27 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, decrement_queue_count(dqm, qpd, q); if (dqm->dev->kfd->shared_resources.enable_mes) { - int err; - - err = remove_queue_mes(dqm, q, qpd); - if (err) { + retval = remove_queue_mes(dqm, q, qpd); + if (retval) { dev_err(dev, "Failed to evict queue %d\n", q->properties.queue_id); - retval = err; + goto out; } } } - pdd->last_evict_timestamp = get_jiffies_64(); - if (!dqm->dev->kfd->shared_resources.enable_mes) + + if (!dqm->dev->kfd->shared_resources.enable_mes) { + pdd->last_evict_timestamp = get_jiffies_64(); retval = execute_queues_cpsch(dqm, qpd->is_debug ? KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES : KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0, USE_DEFAULT_GRACE_PERIOD); + } else { + retval = resume_all_queues_mes(dqm); + if (retval) + dev_err(dev, "Resuming all queues failed"); + } out: dqm_unlock(dqm); @@ -1884,6 +1897,8 @@ fail_packet_manager_init: static int stop_cpsch(struct device_queue_manager *dqm) { + int ret = 0; + dqm_lock(dqm); if (!dqm->sched_running) { dqm_unlock(dqm); @@ -1891,9 +1906,10 @@ static int stop_cpsch(struct device_queue_manager *dqm) } if (!dqm->dev->kfd->shared_resources.enable_mes) - unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, USE_DEFAULT_GRACE_PERIOD, false); + ret = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, + 0, USE_DEFAULT_GRACE_PERIOD, false); else - remove_all_kfd_queues_mes(dqm); + ret = remove_all_kfd_queues_mes(dqm); dqm->sched_running = false; @@ -1907,7 +1923,7 @@ static int stop_cpsch(struct device_queue_manager *dqm) dqm->detect_hang_info = NULL; dqm_unlock(dqm); - return 0; + return ret; } static int create_kernel_queue_cpsch(struct device_queue_manager *dqm, @@ -2078,7 +2094,8 @@ int amdkfd_fence_wait_timeout(struct device_queue_manager *dqm, while (*fence_addr != fence_value) { /* Fatal err detected, this response won't come */ - if (amdgpu_amdkfd_is_fed(dqm->dev->adev)) + if (amdgpu_amdkfd_is_fed(dqm->dev->adev) || + amdgpu_in_reset(dqm->dev->adev)) return -EIO; if (time_after(jiffies, end_jiffies)) { @@ -3098,61 +3115,17 @@ out: return ret; } -static int kfd_dqm_evict_pasid_mes(struct device_queue_manager *dqm, - struct qcm_process_device *qpd) -{ - struct device *dev = dqm->dev->adev->dev; - int ret = 0; - - /* Check if process is already evicted */ - dqm_lock(dqm); - if (qpd->evicted) { - /* Increment the evicted count to make sure the - * process stays evicted before its terminated. - */ - qpd->evicted++; - dqm_unlock(dqm); - goto out; - } - dqm_unlock(dqm); - - ret = suspend_all_queues_mes(dqm); - if (ret) { - dev_err(dev, "Suspending all queues failed"); - goto out; - } - - ret = dqm->ops.evict_process_queues(dqm, qpd); - if (ret) { - dev_err(dev, "Evicting process queues failed"); - goto out; - } - - ret = resume_all_queues_mes(dqm); - if (ret) - dev_err(dev, "Resuming all queues failed"); - -out: - return ret; -} - int kfd_evict_process_device(struct kfd_process_device *pdd) { struct device_queue_manager *dqm; struct kfd_process *p; - int ret = 0; p = pdd->process; dqm = pdd->dev->dqm; WARN(debug_evictions, "Evicting pid %d", p->lead_thread->pid); - if (dqm->dev->kfd->shared_resources.enable_mes) - ret = kfd_dqm_evict_pasid_mes(dqm, &pdd->qpd); - else - ret = dqm->ops.evict_process_queues(dqm, &pdd->qpd); - - return ret; + return dqm->ops.evict_process_queues(dqm, &pdd->qpd); } int reserve_debug_trap_vmid(struct device_queue_manager *dqm, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c index 82905f3e54dd..5a190dd6be4e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -748,16 +748,6 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id, uint64_t *slots = page_slots(p->signal_page); uint32_t id; - /* - * If id is valid but slot is not signaled, GPU may signal the same event twice - * before driver have chance to process the first interrupt, then signal slot is - * auto-reset after set_event wakeup the user space, just drop the second event as - * the application only need wakeup once. - */ - if ((valid_id_bits > 31 || (1U << valid_id_bits) >= KFD_SIGNAL_EVENT_LIMIT) && - partial_id < KFD_SIGNAL_EVENT_LIMIT && slots[partial_id] == UNSIGNALED_EVENT_SLOT) - goto out_unlock; - if (valid_id_bits) pr_debug_ratelimited("Partial ID invalid: %u (%u valid bits)\n", partial_id, valid_id_bits); @@ -786,7 +776,6 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id, } } -out_unlock: rcu_read_unlock(); kfd_unref_process(p); } 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 4ceb251312a6..d76fb61869c7 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -28,6 +28,7 @@ #include "kfd_device_queue_manager.h" #include "kfd_smi_events.h" #include "amdgpu_ras.h" +#include "amdgpu_ras_mgr.h" /* * GFX9 SQ Interrupts @@ -228,7 +229,11 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, kfd_signal_poison_consumed_event(dev, pasid); - event_id = amdgpu_ras_acquire_event_id(dev->adev, type); + if (amdgpu_uniras_enabled(dev->adev)) + event_id = amdgpu_ras_mgr_gen_ras_event_seqno(dev->adev, + RAS_SEQNO_TYPE_POISON_CONSUMPTION); + else + event_id = amdgpu_ras_acquire_event_id(dev->adev, type); RAS_EVENT_LOG(dev->adev, event_id, "poison is consumed by client %d, kick off gpu reset flow\n", client_id); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 59a5a3fea65d..46c84fc60af1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -21,7 +21,6 @@ * OTHER DEALINGS IN THE SOFTWARE. */ #include <linux/types.h> -#include <linux/hmm.h> #include <linux/dma-direction.h> #include <linux/dma-mapping.h> #include <linux/migrate.h> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h index 2eebf67f9c2c..2b7fd442d29c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h @@ -31,7 +31,6 @@ #include <linux/list.h> #include <linux/mutex.h> #include <linux/sched/mm.h> -#include <linux/hmm.h> #include "kfd_priv.h" #include "kfd_svm.h" diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index ddfe30c13e9d..a085faac9fe1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -1083,7 +1083,6 @@ static void kfd_process_destroy_pdds(struct kfd_process *p) * for auto suspend */ if (pdd->runtime_inuse) { - pm_runtime_mark_last_busy(adev_to_drm(pdd->dev->adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(pdd->dev->adev)->dev); pdd->runtime_inuse = false; } @@ -1162,9 +1161,6 @@ static void kfd_process_wq_release(struct work_struct *work) release_work); struct dma_fence *ef; - kfd_process_dequeue_from_all_devices(p); - pqm_uninit(&p->pqm); - /* * If GPU in reset, user queues may still running, wait for reset complete. */ @@ -1226,6 +1222,14 @@ static void kfd_process_notifier_release_internal(struct kfd_process *p) cancel_delayed_work_sync(&p->eviction_work); cancel_delayed_work_sync(&p->restore_work); + /* + * Dequeue and destroy user queues, it is not safe for GPU to access + * system memory after mmu release notifier callback returns because + * exit_mmap free process memory afterwards. + */ + kfd_process_dequeue_from_all_devices(p); + pqm_uninit(&p->pqm); + for (i = 0; i < p->n_pdds; i++) { struct kfd_process_device *pdd = p->pdds[i]; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c index a65c67cf56ff..f1e7583650c4 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c @@ -297,16 +297,16 @@ int kfd_queue_acquire_buffers(struct kfd_process_device *pdd, struct queue_prope goto out_err_unreserve; } - if (properties->ctx_save_restore_area_size != topo_dev->node_props.cwsr_size) { - pr_debug("queue cwsr size 0x%x not equal to node cwsr size 0x%x\n", + if (properties->ctx_save_restore_area_size < topo_dev->node_props.cwsr_size) { + pr_debug("queue cwsr size 0x%x not sufficient for node cwsr size 0x%x\n", properties->ctx_save_restore_area_size, topo_dev->node_props.cwsr_size); err = -EINVAL; goto out_err_unreserve; } - total_cwsr_size = (topo_dev->node_props.cwsr_size + topo_dev->node_props.debug_memory_size) - * NUM_XCC(pdd->dev->xcc_mask); + total_cwsr_size = (properties->ctx_save_restore_area_size + + topo_dev->node_props.debug_memory_size) * NUM_XCC(pdd->dev->xcc_mask); total_cwsr_size = ALIGN(total_cwsr_size, PAGE_SIZE); err = kfd_queue_buffer_get(vm, (void *)properties->ctx_save_restore_area_address, @@ -352,8 +352,8 @@ int kfd_queue_release_buffers(struct kfd_process_device *pdd, struct queue_prope topo_dev = kfd_topology_device_by_id(pdd->dev->id); if (!topo_dev) return -EINVAL; - total_cwsr_size = (topo_dev->node_props.cwsr_size + topo_dev->node_props.debug_memory_size) - * NUM_XCC(pdd->dev->xcc_mask); + total_cwsr_size = (properties->ctx_save_restore_area_size + + topo_dev->node_props.debug_memory_size) * NUM_XCC(pdd->dev->xcc_mask); total_cwsr_size = ALIGN(total_cwsr_size, PAGE_SIZE); kfd_queue_buffer_svm_put(pdd, properties->ctx_save_restore_area_address, total_cwsr_size); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 9d72411c3379..97c2270f278f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1698,7 +1698,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm, start = map_start << PAGE_SHIFT; end = (map_last + 1) << PAGE_SHIFT; for (addr = start; !r && addr < end; ) { - struct hmm_range *hmm_range = NULL; + struct amdgpu_hmm_range *range = NULL; unsigned long map_start_vma; unsigned long map_last_vma; struct vm_area_struct *vma; @@ -1737,9 +1737,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm, } WRITE_ONCE(p->svms.faulting_task, current); - r = amdgpu_hmm_range_get_pages(&prange->notifier, addr, npages, - readonly, owner, - &hmm_range); + range = amdgpu_hmm_range_alloc(NULL); + if (likely(range)) + r = amdgpu_hmm_range_get_pages(&prange->notifier, addr, npages, + readonly, owner, range); + else + r = -ENOMEM; WRITE_ONCE(p->svms.faulting_task, NULL); if (r) pr_debug("failed %d to get svm range pages\n", r); @@ -1750,7 +1753,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm, if (!r) { offset = (addr >> PAGE_SHIFT) - prange->start; r = svm_range_dma_map(prange, ctx->bitmap, offset, npages, - hmm_range->hmm_pfns); + range->hmm_range.hmm_pfns); if (r) pr_debug("failed %d to dma map range\n", r); } @@ -1758,14 +1761,17 @@ static int svm_range_validate_and_map(struct mm_struct *mm, svm_range_lock(prange); /* Free backing memory of hmm_range if it was initialized - * Overrride return value to TRY AGAIN only if prior returns + * Override return value to TRY AGAIN only if prior returns * were successful */ - if (hmm_range && amdgpu_hmm_range_get_pages_done(hmm_range) && !r) { + if (range && !amdgpu_hmm_range_valid(range) && !r) { pr_debug("hmm update the range, need validate again\n"); r = -EAGAIN; } + /* Free the hmm range */ + amdgpu_hmm_range_free(range); + if (!r && !list_empty(&prange->child_list)) { pr_debug("range split by unmap in parallel, validate again\n"); r = -EAGAIN; @@ -3687,6 +3693,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping); /* TODO: unmap ranges from GPU that lost access */ } + update_mapping |= !p->xnack_enabled && !list_empty(&remap_list); + list_for_each_entry_safe(prange, next, &remove_list, update_list) { pr_debug("unlink old 0x%p prange 0x%p [0x%lx 0x%lx]\n", prange->svms, prange, prange->start, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h index 01c7a4877904..a63dfc95b602 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -31,7 +31,6 @@ #include <linux/list.h> #include <linux/mutex.h> #include <linux/sched/mm.h> -#include <linux/hmm.h> #include "amdgpu.h" #include "kfd_priv.h" diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 5c98746eb72d..811636af14ea 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -530,7 +530,9 @@ 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 && + (dev->gpu->xcp->xcp_mgr->mode != + AMDGPU_SPX_PARTITION_MODE) ? dev->gpu->xcp->unique_id : dev->gpu->adev->unique_id); sysfs_show_32bit_prop(buffer, offs, "num_xcc", |