diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 7 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 52 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 6 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 156 |
7 files changed, 163 insertions, 64 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index 0fffaf859c59..3b119db16003 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -406,7 +406,7 @@ static const struct kfd_device_info aldebaran_device_info = { static const struct kfd_device_info renoir_device_info = { .asic_family = CHIP_RENOIR, .asic_name = "renoir", - .gfx_target_version = 90002, + .gfx_target_version = 90012, .max_pasid_bits = 16, .max_no_of_hqd = 24, .doorbell_size = 8, 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 533b27b35fc9..93e33dd84dd4 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -1226,6 +1226,11 @@ static int stop_cpsch(struct device_queue_manager *dqm) bool hanging; dqm_lock(dqm); + if (!dqm->sched_running) { + dqm_unlock(dqm); + return 0; + } + if (!dqm->is_hws_hang) unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0); hanging = dqm->is_hws_hang || dqm->is_resetting; @@ -1430,7 +1435,7 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm, if (!dqm->sched_running) return 0; - if (dqm->is_hws_hang) + if (dqm->is_hws_hang || dqm->is_resetting) return -EIO; if (!dqm->active_runlist) return retval; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c index 2e86692def19..d1388896f9c1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c @@ -308,7 +308,7 @@ * 16MB are reserved for kernel use (CWSR trap handler and kernel IB * for now). */ -#define SVM_USER_BASE 0x1000000ull +#define SVM_USER_BASE (u64)(KFD_CWSR_TBA_TMA_SIZE + 2*PAGE_SIZE) #define SVM_CWSR_BASE (SVM_USER_BASE - KFD_CWSR_TBA_TMA_SIZE) #define SVM_IB_BASE (SVM_CWSR_BASE - PAGE_SIZE) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 6d8634e40b3b..9b9c2b9bf2ef 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -281,6 +281,19 @@ static unsigned long svm_migrate_successful_pages(struct migrate_vma *migrate) return cpages; } +static unsigned long svm_migrate_unsuccessful_pages(struct migrate_vma *migrate) +{ + unsigned long upages = 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++; + } + return upages; +} + static int svm_migrate_copy_to_vram(struct amdgpu_device *adev, struct svm_range *prange, struct migrate_vma *migrate, struct dma_fence **mfence, @@ -317,7 +330,6 @@ svm_migrate_copy_to_vram(struct amdgpu_device *adev, struct svm_range *prange, migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); svm_migrate_get_vram_page(prange, migrate->dst[i]); migrate->dst[i] = migrate_pfn(migrate->dst[i]); - migrate->dst[i] |= MIGRATE_PFN_LOCKED; src[i] = dma_map_page(dev, spage, 0, PAGE_SIZE, DMA_TO_DEVICE); r = dma_mapping_error(dev, src[i]); @@ -610,7 +622,6 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, dst[i] >> PAGE_SHIFT, page_to_pfn(dpage)); migrate->dst[i] = migrate_pfn(page_to_pfn(dpage)); - migrate->dst[i] |= MIGRATE_PFN_LOCKED; j++; } @@ -634,10 +645,11 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, struct vm_area_struct *vma, uint64_t start, uint64_t end) { uint64_t npages = (end - start) >> PAGE_SHIFT; + unsigned long upages = npages; + unsigned long cpages = 0; struct kfd_process_device *pdd; struct dma_fence *mfence = NULL; struct migrate_vma migrate; - unsigned long cpages = 0; dma_addr_t *scratch; size_t size; void *buf; @@ -671,6 +683,7 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, 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) @@ -683,8 +696,9 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, scratch, npages); migrate_vma_pages(&migrate); - pr_debug("successful/cpages/npages 0x%lx/0x%lx/0x%lx\n", - svm_migrate_successful_pages(&migrate), cpages, migrate.npages); + upages = svm_migrate_unsuccessful_pages(&migrate); + pr_debug("unsuccessful/cpages/npages 0x%lx/0x%lx/0x%lx\n", + upages, cpages, migrate.npages); svm_migrate_copy_done(adev, mfence); migrate_vma_finalize(&migrate); @@ -698,9 +712,9 @@ out: if (pdd) WRITE_ONCE(pdd->page_out, pdd->page_out + cpages); - return cpages; + return upages; } - return r; + return r ? r : upages; } /** @@ -720,7 +734,7 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm) unsigned long addr; unsigned long start; unsigned long end; - unsigned long cpages = 0; + unsigned long upages = 0; long r = 0; if (!prange->actual_loc) { @@ -756,12 +770,12 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm) pr_debug("failed %ld to migrate\n", r); break; } else { - cpages += r; + upages += r; } addr = next; } - if (cpages) { + if (!upages) { svm_range_vram_node_free(prange); prange->actual_loc = 0; } @@ -784,7 +798,7 @@ static int svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, struct mm_struct *mm) { - int r; + int r, retries = 3; /* * TODO: for both devices with PCIe large bar or on same xgmi hive, skip @@ -793,9 +807,14 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc, pr_debug("from gpu 0x%x to gpu 0x%x\n", prange->actual_loc, best_loc); - r = svm_migrate_vram_to_ram(prange, mm); - if (r) - return r; + do { + r = svm_migrate_vram_to_ram(prange, mm); + if (r) + return r; + } while (prange->actual_loc && --retries); + + if (prange->actual_loc) + return -EDEADLK; return svm_migrate_ram_to_vram(prange, best_loc, mm); } @@ -840,6 +859,11 @@ static vm_fault_t svm_migrate_to_ram(struct vm_fault *vmf) pr_debug("failed find process at fault address 0x%lx\n", addr); return VM_FAULT_SIGBUS; } + if (READ_ONCE(p->svms.faulting_task) == current) { + pr_debug("skipping ram migration\n"); + kfd_unref_process(p); + return 0; + } addr >>= PAGE_SHIFT; pr_debug("CPU page fault svms 0x%p address 0x%lx\n", &p->svms, addr); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 4104b167e721..8fd48d0ed240 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -766,8 +766,10 @@ struct svm_range_list { struct list_head deferred_range_list; spinlock_t deferred_list_lock; atomic_t evicted_ranges; + atomic_t drain_pagefaults; struct delayed_work restore_work; DECLARE_BITMAP(bitmap_supported, MAX_GPU_INSTANCE); + struct task_struct *faulting_task; }; /* Process data */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 457863861d6f..b993011cfa64 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -1715,7 +1715,11 @@ int kfd_process_evict_queues(struct kfd_process *p) r = pdd->dev->dqm->ops.evict_process_queues(pdd->dev->dqm, &pdd->qpd); - if (r) { + /* evict return -EIO if HWS is hang or asic is resetting, in this case + * we would like to set all the queues to be in evicted state to prevent + * them been add back since they actually not be saved right now. + */ + if (r && r != -EIO) { pr_err("Failed to evict process queues\n"); goto fail; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index b691c8495d66..3cb4681c5f53 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1496,9 +1496,11 @@ static int svm_range_validate_and_map(struct mm_struct *mm, next = min(vma->vm_end, end); npages = (next - addr) >> PAGE_SHIFT; + WRITE_ONCE(p->svms.faulting_task, current); r = amdgpu_hmm_range_get_pages(&prange->notifier, mm, NULL, addr, npages, &hmm_range, readonly, true, owner); + WRITE_ONCE(p->svms.faulting_task, NULL); if (r) { pr_debug("failed %d to get svm range pages\n", r); goto unreserve_out; @@ -1572,7 +1574,6 @@ retry_flush_work: static void svm_range_restore_work(struct work_struct *work) { struct delayed_work *dwork = to_delayed_work(work); - struct amdkfd_process_info *process_info; struct svm_range_list *svms; struct svm_range *prange; struct kfd_process *p; @@ -1592,12 +1593,10 @@ static void svm_range_restore_work(struct work_struct *work) * the lifetime of this thread, kfd_process and mm will be valid. */ p = container_of(svms, struct kfd_process, svms); - process_info = p->kgd_process_info; mm = p->mm; if (!mm) return; - mutex_lock(&process_info->lock); svm_range_list_lock_and_flush_work(svms, mm); mutex_lock(&svms->lock); @@ -1650,7 +1649,6 @@ static void svm_range_restore_work(struct work_struct *work) out_reschedule: mutex_unlock(&svms->lock); mmap_write_unlock(mm); - mutex_unlock(&process_info->lock); /* If validation failed, reschedule another attempt */ if (evicted_ranges) { @@ -1966,10 +1964,16 @@ 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; p = container_of(svms, struct kfd_process, svms); +restart: + drain = atomic_read(&svms->drain_pagefaults); + if (!drain) + return; + for_each_set_bit(i, svms->bitmap_supported, p->n_pdds) { pdd = p->pdds[i]; if (!pdd) @@ -1981,6 +1985,8 @@ static void svm_range_drain_retry_fault(struct svm_range_list *svms) amdgpu_ih_wait_on_checkpoint_process(adev, &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) + goto restart; } static void svm_range_deferred_list_work(struct work_struct *work) @@ -1988,35 +1994,41 @@ static void svm_range_deferred_list_work(struct work_struct *work) struct svm_range_list *svms; struct svm_range *prange; struct mm_struct *mm; + struct kfd_process *p; svms = container_of(work, struct svm_range_list, deferred_list_work); pr_debug("enter svms 0x%p\n", svms); + p = container_of(svms, struct kfd_process, svms); + /* Avoid mm is gone when inserting mmu notifier */ + mm = get_task_mm(p->lead_thread); + if (!mm) { + pr_debug("svms 0x%p process mm gone\n", svms); + return; + } +retry: + mmap_write_lock(mm); + + /* Checking for the need to drain retry faults must be inside + * mmap write lock to serialize with munmap notifiers. + */ + if (unlikely(atomic_read(&svms->drain_pagefaults))) { + mmap_write_unlock(mm); + svm_range_drain_retry_fault(svms); + goto retry; + } + spin_lock(&svms->deferred_list_lock); while (!list_empty(&svms->deferred_range_list)) { prange = list_first_entry(&svms->deferred_range_list, struct svm_range, deferred_list); + list_del_init(&prange->deferred_list); spin_unlock(&svms->deferred_list_lock); + pr_debug("prange 0x%p [0x%lx 0x%lx] op %d\n", prange, prange->start, prange->last, prange->work_item.op); - /* Make sure no stale retry fault coming after range is freed */ - if (prange->work_item.op == SVM_OP_UNMAP_RANGE) - svm_range_drain_retry_fault(prange->svms); - - mm = prange->work_item.mm; - mmap_write_lock(mm); mutex_lock(&svms->lock); - - /* Remove from deferred_list must be inside mmap write lock, - * otherwise, svm_range_list_lock_and_flush_work may hold mmap - * write lock, and continue because deferred_list is empty, then - * deferred_list handle is blocked by mmap write lock. - */ - spin_lock(&svms->deferred_list_lock); - list_del_init(&prange->deferred_list); - spin_unlock(&svms->deferred_list_lock); - mutex_lock(&prange->migrate_mutex); while (!list_empty(&prange->child_list)) { struct svm_range *pchild; @@ -2032,12 +2044,13 @@ static void svm_range_deferred_list_work(struct work_struct *work) svm_range_handle_list_op(svms, prange); mutex_unlock(&svms->lock); - mmap_write_unlock(mm); spin_lock(&svms->deferred_list_lock); } spin_unlock(&svms->deferred_list_lock); + mmap_write_unlock(mm); + mmput(mm); pr_debug("exit svms 0x%p\n", svms); } @@ -2124,6 +2137,12 @@ svm_range_unmap_from_cpu(struct mm_struct *mm, struct svm_range *prange, pr_debug("svms 0x%p prange 0x%p [0x%lx 0x%lx] [0x%lx 0x%lx]\n", svms, prange, prange->start, prange->last, start, last); + /* Make sure pending page faults are drained in the deferred worker + * before the range is freed to avoid straggler interrupts on + * unmapped memory causing "phantom faults". + */ + atomic_inc(&svms->drain_pagefaults); + unmap_parent = start <= prange->start && last >= prange->last; list_for_each_entry(pchild, &prange->child_list, child_list) { @@ -2261,7 +2280,7 @@ svm_range_from_addr(struct svm_range_list *svms, unsigned long addr, * migration if actual loc is not best location, then update GPU page table * mapping to the best location. * - * If vm fault gpu is range preferred loc, the best_loc is preferred loc. + * If the preferred loc is accessible by faulting GPU, use preferred loc. * If vm fault gpu idx is on range ACCESSIBLE bitmap, best_loc is vm fault gpu * If vm fault gpu idx is on range ACCESSIBLE_IN_PLACE bitmap, then * if range actual loc is cpu, best_loc is cpu @@ -2278,7 +2297,7 @@ svm_range_best_restore_location(struct svm_range *prange, struct amdgpu_device *adev, int32_t *gpuidx) { - struct amdgpu_device *bo_adev; + struct amdgpu_device *bo_adev, *preferred_adev; struct kfd_process *p; uint32_t gpuid; int r; @@ -2291,8 +2310,16 @@ svm_range_best_restore_location(struct svm_range *prange, return -1; } - if (prange->preferred_loc == gpuid) + if (prange->preferred_loc == gpuid || + prange->preferred_loc == KFD_IOCTL_SVM_LOCATION_SYSMEM) { return prange->preferred_loc; + } else if (prange->preferred_loc != KFD_IOCTL_SVM_LOCATION_UNDEFINED) { + preferred_adev = svm_range_get_adev_by_id(prange, + prange->preferred_loc); + if (amdgpu_xgmi_same_hive(adev, preferred_adev)) + return prange->preferred_loc; + /* fall through */ + } if (test_bit(*gpuidx, prange->bitmap_access)) return gpuid; @@ -2313,7 +2340,8 @@ svm_range_best_restore_location(struct svm_range *prange, static int svm_range_get_range_boundaries(struct kfd_process *p, int64_t addr, - unsigned long *start, unsigned long *last) + unsigned long *start, unsigned long *last, + bool *is_heap_stack) { struct vm_area_struct *vma; struct interval_tree_node *node; @@ -2324,6 +2352,12 @@ svm_range_get_range_boundaries(struct kfd_process *p, int64_t addr, pr_debug("VMA does not exist in address [0x%llx]\n", addr); return -EFAULT; } + + *is_heap_stack = (vma->vm_start <= vma->vm_mm->brk && + vma->vm_end >= vma->vm_mm->start_brk) || + (vma->vm_start <= vma->vm_mm->start_stack && + vma->vm_end >= vma->vm_mm->start_stack); + start_limit = max(vma->vm_start >> PAGE_SHIFT, (unsigned long)ALIGN_DOWN(addr, 2UL << 8)); end_limit = min(vma->vm_end >> PAGE_SHIFT, @@ -2353,9 +2387,9 @@ svm_range_get_range_boundaries(struct kfd_process *p, int64_t addr, *start = start_limit; *last = end_limit - 1; - pr_debug("vma start: 0x%lx start: 0x%lx vma end: 0x%lx last: 0x%lx\n", - vma->vm_start >> PAGE_SHIFT, *start, - vma->vm_end >> PAGE_SHIFT, *last); + pr_debug("vma [0x%lx 0x%lx] range [0x%lx 0x%lx] is_heap_stack %d\n", + vma->vm_start >> PAGE_SHIFT, vma->vm_end >> PAGE_SHIFT, + *start, *last, *is_heap_stack); return 0; } @@ -2420,11 +2454,13 @@ svm_range *svm_range_create_unregistered_range(struct amdgpu_device *adev, struct svm_range *prange = NULL; unsigned long start, last; uint32_t gpuid, gpuidx; + bool is_heap_stack; uint64_t bo_s = 0; uint64_t bo_l = 0; int r; - if (svm_range_get_range_boundaries(p, addr, &start, &last)) + if (svm_range_get_range_boundaries(p, addr, &start, &last, + &is_heap_stack)) return NULL; r = svm_range_check_vm(p, start, last, &bo_s, &bo_l); @@ -2451,6 +2487,9 @@ svm_range *svm_range_create_unregistered_range(struct amdgpu_device *adev, return NULL; } + if (is_heap_stack) + prange->preferred_loc = KFD_IOCTL_SVM_LOCATION_SYSMEM; + svm_range_add_to_svms(prange); svm_range_add_notifier_locked(mm, prange); @@ -2523,20 +2562,13 @@ svm_range_count_fault(struct amdgpu_device *adev, struct kfd_process *p, } static bool -svm_fault_allowed(struct mm_struct *mm, uint64_t addr, bool write_fault) +svm_fault_allowed(struct vm_area_struct *vma, bool write_fault) { unsigned long requested = VM_READ; - struct vm_area_struct *vma; if (write_fault) requested |= VM_WRITE; - vma = find_vma(mm, addr << PAGE_SHIFT); - if (!vma || (addr << PAGE_SHIFT) < vma->vm_start) { - pr_debug("address 0x%llx VMA is removed\n", addr); - return true; - } - pr_debug("requested 0x%lx, vma permission flags 0x%lx\n", requested, vma->vm_flags); return (vma->vm_flags & requested) == requested; @@ -2554,6 +2586,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, int32_t best_loc; int32_t gpuidx = MAX_GPU_INSTANCE; bool write_locked = false; + struct vm_area_struct *vma; int r = 0; if (!KFD_IS_SVM_API_SUPPORTED(adev->kfd.dev)) { @@ -2564,7 +2597,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, p = kfd_lookup_process_by_pasid(pasid); if (!p) { pr_debug("kfd process not founded pasid 0x%x\n", pasid); - return -ESRCH; + return 0; } if (!p->xnack_enabled) { pr_debug("XNACK not enabled for pasid 0x%x\n", pasid); @@ -2575,10 +2608,19 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, pr_debug("restoring svms 0x%p fault address 0x%llx\n", svms, addr); + if (atomic_read(&svms->drain_pagefaults)) { + pr_debug("draining retry fault, drop fault 0x%llx\n", addr); + r = 0; + goto out; + } + + /* p->lead_thread is available as kfd_process_wq_release flush the work + * before releasing task ref. + */ mm = get_task_mm(p->lead_thread); if (!mm) { pr_debug("svms 0x%p failed to get mm\n", svms); - r = -ESRCH; + r = 0; goto out; } @@ -2616,6 +2658,7 @@ retry_write_locked: if (svm_range_skip_recover(prange)) { amdgpu_gmc_filter_faults_remove(adev, addr, pasid); + r = 0; goto out_unlock_range; } @@ -2624,10 +2667,21 @@ retry_write_locked: if (timestamp < AMDGPU_SVM_RANGE_RETRY_FAULT_PENDING) { pr_debug("svms 0x%p [0x%lx %lx] already restored\n", svms, prange->start, prange->last); + r = 0; + goto out_unlock_range; + } + + /* __do_munmap removed VMA, return success as we are handling stale + * retry fault. + */ + vma = find_vma(mm, addr << PAGE_SHIFT); + if (!vma || (addr << PAGE_SHIFT) < vma->vm_start) { + pr_debug("address 0x%llx VMA is removed\n", addr); + r = 0; goto out_unlock_range; } - if (!svm_fault_allowed(mm, addr, write_fault)) { + if (!svm_fault_allowed(vma, write_fault)) { pr_debug("fault addr 0x%llx no %s permission\n", addr, write_fault ? "write" : "read"); r = -EPERM; @@ -2705,6 +2759,14 @@ void svm_range_list_fini(struct kfd_process *p) /* Ensure list work is finished before process is destroyed */ flush_work(&p->svms.deferred_list_work); + /* + * Ensure no retry fault comes in afterwards, as page fault handler will + * not find kfd process and take mm lock to recover fault. + */ + atomic_inc(&p->svms.drain_pagefaults); + svm_range_drain_retry_fault(&p->svms); + + list_for_each_entry_safe(prange, next, &p->svms.list, list) { svm_range_unlink(prange); svm_range_remove_notifier(prange); @@ -2725,6 +2787,7 @@ int svm_range_list_init(struct kfd_process *p) mutex_init(&svms->lock); INIT_LIST_HEAD(&svms->list); atomic_set(&svms->evicted_ranges, 0); + atomic_set(&svms->drain_pagefaults, 0); INIT_DELAYED_WORK(&svms->restore_work, svm_range_restore_work); INIT_WORK(&svms->deferred_list_work, svm_range_deferred_list_work); INIT_LIST_HEAD(&svms->deferred_range_list); @@ -3076,6 +3139,8 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work) struct svm_range *prange = list_first_entry(&svm_bo->range_list, struct svm_range, svm_bo_list); + int retries = 3; + list_del_init(&prange->svm_bo_list); spin_unlock(&svm_bo->list_lock); @@ -3083,7 +3148,11 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work) prange->start, prange->last); mutex_lock(&prange->migrate_mutex); - svm_migrate_vram_to_ram(prange, svm_bo->eviction_fence->mm); + do { + svm_migrate_vram_to_ram(prange, + svm_bo->eviction_fence->mm); + } while (prange->actual_loc && --retries); + WARN(prange->actual_loc, "Migration failed during eviction"); mutex_lock(&prange->lock); prange->svm_bo = NULL; @@ -3108,7 +3177,6 @@ static int svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs) { - struct amdkfd_process_info *process_info = p->kgd_process_info; struct mm_struct *mm = current->mm; struct list_head update_list; struct list_head insert_list; @@ -3127,8 +3195,6 @@ svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, svms = &p->svms; - mutex_lock(&process_info->lock); - svm_range_list_lock_and_flush_work(svms, mm); r = svm_range_is_valid(p, start, size); @@ -3204,8 +3270,6 @@ out_unlock_range: mutex_unlock(&svms->lock); mmap_read_unlock(mm); out: - mutex_unlock(&process_info->lock); - pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n", p->pasid, &p->svms, start, start + size - 1, r); |