drm/amdkfd: Change kfd/svm page fault drain handling
Subsystems (3)
PATCH v4 1/1 drm/amdkfd: Change kfd/svm page fault drain handling
From: Xiaogang Chen <[email protected]>
When app unmap vm ranges(munmap) kfd/svm starts drain pending page fault and not handle any incoming pages fault of this process until a deferred work item got executed by default system wq. The time period of "not handle page fault" can be long and is unpredicable. That is advese to kfd performance on page faults recovery.
This patch uses time stamp of incoming page page to decide to drop or handle page fault. When app unmap vm ranges kfd records each gpu device's ih ring current time stamp. These time stamps are used at kfd page fault recovery routine.
Any page fault happens on unmapped ranges after unmap events is app bug that accesses vm range after unmap. It is not driver work to cover that.
By using time stamp of page fault do not need drain page faults at deferred work. So, the time period that kfd does not handle page faults is reduced and can be controlled.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 4 +- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 +- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 128 +++++++++++-------------- drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 2 +- 7 files changed, 68 insertions(+), 78 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +2 -2
@@ -2763,7 +2763,7 @@ int amdgpu_vm_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)* shouldn't be reported any more. */ bool amdgpu_vm_handle_fault (struct amdgpu_device *adev,u32 pasid, -u32 vmid,u32 node_id,uint64_t addr, +u32 vmid,u32 node_id,uint64_t addr,uint64_t ts,bool write_fault) {bool is_compute_context = false;
@@ -2789,7 +2789,7 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid, addr /=AMDGPU_GPU_PAGE_SIZE ;if (is_compute_context && !svm_range_restore_pages (adev, pasid, vmid, - node_id, addr, write_fault)) { + node_id, addr, ts, write_fault)) {amdgpu_bo_unref (&root);return true; }
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +1 -1
@@ -548,7 +548,7 @@ amdgpu_vm_get_task_info_vm(struct amdgpu_vm *vm);void amdgpu_vm_put_task_info (struct amdgpu_task_info *task_info);bool amdgpu_vm_handle_fault (struct amdgpu_device *adev,u32 pasid, -u32 vmid,u32 node_id,uint64_t addr, +u32 vmid,u32 node_id,uint64_t addr,uint64_t ts,bool write_fault);void amdgpu_vm_set_task_info (struct amdgpu_vm *vm);
drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +2 -1
@@ -132,7 +132,8 @@ static int gmc_v10_0_process_interrupt(struct amdgpu_device *adev,/* Try to handle the recoverable page faults by filling page * tables */ -if (amdgpu_vm_handle_fault (adev, entry->pasid,0 ,0 , addr, write_fault)) +if (amdgpu_vm_handle_fault (adev, entry->pasid,0 ,0 , addr, + entry->timestamp, write_fault))return 1 ; }
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +2 -2
@@ -595,7 +595,7 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, cam_index = entry->src_data[2 ] &0x3ff ; ret =amdgpu_vm_handle_fault (adev, entry->pasid, entry->vmid, node_id, - addr, write_fault); + addr, entry->timestamp, write_fault);WDOORBELL32 (adev->irq.retry_cam_doorbell_index, cam_index);if (ret)return 1 ;
@@ -618,7 +618,7 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,* tables */ if (amdgpu_vm_handle_fault (adev, entry->pasid, entry->vmid, node_id, - addr, write_fault)) + addr, entry->timestamp, write_fault))return 1 ; } }
drivers/gpu/drm/amd/amdkfd/kfd_priv.h +2 -1
@@ -850,10 +850,11 @@ struct svm_range_list {struct list_head criu_svm_metadata_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; +/* check point ts decides if page fault recovery need be dropped */ +uint64_t checkpoint_ts[MAX_GPU_INSTANCE ]; };/* Process data */
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +58 -70
193 lines not on this page · raw message
drivers/gpu/drm/amd/amdkfd/kfd_svm.h +1 -1
@@ -173,7 +173,7 @@ int svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange, bool clear);void svm_range_vram_node_free (struct svm_range *prange);int svm_range_restore_pages (struct amdgpu_device *adev,unsigned int pasid, -uint32_t vmid,uint32_t node_id,uint64_t addr, +uint32_t vmid,uint32_t node_id,uint64_t addr,uint64_t ts,bool write_fault);int svm_range_schedule_evict_svm_bo (struct amdgpu_amdkfd_fence *fence); void svm_range_add_list_work(struct svm_range_list *svms,
Signature · 1 line
2.25.1