| 1 | // SPDX-License-Identifier: GPL-2.0 OR MIT |
| 2 | /* |
| 3 | * Copyright 2023 Advanced Micro Devices, Inc. |
| 4 | * |
| 5 | * Permission is hereby granted, free of charge, to any person obtaining a |
| 6 | * copy of this software and associated documentation files (the "Software"), |
| 7 | * to deal in the Software without restriction, including without limitation |
| 8 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 9 | * and/or sell copies of the Software, and to permit persons to whom the |
| 10 | * Software is furnished to do so, subject to the following conditions: |
| 11 | * |
| 12 | * The above copyright notice and this permission notice shall be included in |
| 13 | * all copies or substantial portions of the Software. |
| 14 | * |
| 15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 18 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 19 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 20 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 21 | * OTHER DEALINGS IN THE SOFTWARE. |
| 22 | */ |
| 23 | |
| 24 | #include <linux/dma-fence.h> |
| 25 | #include <linux/workqueue.h> |
| 26 | |
| 27 | #include "amdgpu.h" |
| 28 | #include "amdgpu_vm.h" |
| 29 | #include "amdgpu_gmc.h" |
| 30 | |
| 31 | struct amdgpu_tlb_fence { |
| 32 | struct dma_fence base; |
| 33 | struct amdgpu_device *adev; |
| 34 | struct dma_fence *dependency; |
| 35 | struct work_struct work; |
| 36 | spinlock_t lock; |
| 37 | uint16_t pasid; |
| 38 | |
| 39 | }; |
| 40 | |
| 41 | static const char *amdgpu_tlb_fence_get_driver_name(struct dma_fence *fence) |
| 42 | { |
| 43 | return "amdgpu tlb fence" ; |
| 44 | } |
| 45 | |
| 46 | static const char *amdgpu_tlb_fence_get_timeline_name(struct dma_fence *f) |
| 47 | { |
| 48 | return "amdgpu tlb timeline" ; |
| 49 | } |
| 50 | |
| 51 | static void amdgpu_tlb_fence_work(struct work_struct *work) |
| 52 | { |
| 53 | struct amdgpu_tlb_fence *f = container_of(work, typeof(*f), work); |
| 54 | int r; |
| 55 | |
| 56 | if (f->dependency) { |
| 57 | dma_fence_wait(fence: f->dependency, intr: false); |
| 58 | dma_fence_put(fence: f->dependency); |
| 59 | f->dependency = NULL; |
| 60 | } |
| 61 | |
| 62 | r = amdgpu_gmc_flush_gpu_tlb_pasid(adev: f->adev, pasid: f->pasid, flush_type: 2, all_hub: true, inst: 0); |
| 63 | if (r) { |
| 64 | dev_err(f->adev->dev, "TLB flush failed for PASID %d.\n" , |
| 65 | f->pasid); |
| 66 | dma_fence_set_error(fence: &f->base, error: r); |
| 67 | } |
| 68 | |
| 69 | dma_fence_signal(fence: &f->base); |
| 70 | dma_fence_put(fence: &f->base); |
| 71 | } |
| 72 | |
| 73 | static const struct dma_fence_ops amdgpu_tlb_fence_ops = { |
| 74 | .get_driver_name = amdgpu_tlb_fence_get_driver_name, |
| 75 | .get_timeline_name = amdgpu_tlb_fence_get_timeline_name |
| 76 | }; |
| 77 | |
| 78 | void amdgpu_vm_tlb_fence_create(struct amdgpu_device *adev, struct amdgpu_vm *vm, |
| 79 | struct dma_fence **fence) |
| 80 | { |
| 81 | struct amdgpu_tlb_fence *f; |
| 82 | |
| 83 | f = kmalloc(sizeof(*f), GFP_KERNEL); |
| 84 | if (!f) { |
| 85 | /* |
| 86 | * We can't fail since the PDEs and PTEs are already updated, so |
| 87 | * just block for the dependency and execute the TLB flush |
| 88 | */ |
| 89 | if (*fence) |
| 90 | dma_fence_wait(fence: *fence, intr: false); |
| 91 | |
| 92 | amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid: vm->pasid, flush_type: 2, all_hub: true, inst: 0); |
| 93 | *fence = dma_fence_get_stub(); |
| 94 | return; |
| 95 | } |
| 96 | |
| 97 | f->adev = adev; |
| 98 | f->dependency = *fence; |
| 99 | f->pasid = vm->pasid; |
| 100 | INIT_WORK(&f->work, amdgpu_tlb_fence_work); |
| 101 | spin_lock_init(&f->lock); |
| 102 | |
| 103 | dma_fence_init64(fence: &f->base, ops: &amdgpu_tlb_fence_ops, lock: &f->lock, |
| 104 | context: vm->tlb_fence_context, seqno: atomic64_read(v: &vm->tlb_seq)); |
| 105 | |
| 106 | /* TODO: We probably need a separate wq here */ |
| 107 | dma_fence_get(fence: &f->base); |
| 108 | schedule_work(work: &f->work); |
| 109 | |
| 110 | *fence = &f->base; |
| 111 | } |
| 112 | |