1 /* 2 * Copyright 2014 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #include "kfd_mqd_manager.h" 25 #include "amdgpu_amdkfd.h" 26 #include "kfd_device_queue_manager.h" 27 28 /* Mapping queue priority to pipe priority, indexed by queue priority */ 29 int pipe_priority_map[] = { 30 KFD_PIPE_PRIORITY_CS_LOW, 31 KFD_PIPE_PRIORITY_CS_LOW, 32 KFD_PIPE_PRIORITY_CS_LOW, 33 KFD_PIPE_PRIORITY_CS_LOW, 34 KFD_PIPE_PRIORITY_CS_LOW, 35 KFD_PIPE_PRIORITY_CS_LOW, 36 KFD_PIPE_PRIORITY_CS_LOW, 37 KFD_PIPE_PRIORITY_CS_MEDIUM, 38 KFD_PIPE_PRIORITY_CS_MEDIUM, 39 KFD_PIPE_PRIORITY_CS_MEDIUM, 40 KFD_PIPE_PRIORITY_CS_MEDIUM, 41 KFD_PIPE_PRIORITY_CS_HIGH, 42 KFD_PIPE_PRIORITY_CS_HIGH, 43 KFD_PIPE_PRIORITY_CS_HIGH, 44 KFD_PIPE_PRIORITY_CS_HIGH, 45 KFD_PIPE_PRIORITY_CS_HIGH 46 }; 47 48 struct kfd_mem_obj *allocate_hiq_mqd(struct kfd_dev *dev, struct queue_properties *q) 49 { 50 struct kfd_mem_obj *mqd_mem_obj = NULL; 51 52 mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL); 53 if (!mqd_mem_obj) 54 return NULL; 55 56 mqd_mem_obj->gtt_mem = dev->dqm->hiq_sdma_mqd.gtt_mem; 57 mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr; 58 mqd_mem_obj->cpu_ptr = dev->dqm->hiq_sdma_mqd.cpu_ptr; 59 60 return mqd_mem_obj; 61 } 62 63 struct kfd_mem_obj *allocate_sdma_mqd(struct kfd_dev *dev, 64 struct queue_properties *q) 65 { 66 struct kfd_mem_obj *mqd_mem_obj = NULL; 67 uint64_t offset; 68 69 mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL); 70 if (!mqd_mem_obj) 71 return NULL; 72 73 offset = (q->sdma_engine_id * 74 dev->device_info->num_sdma_queues_per_engine + 75 q->sdma_queue_id) * 76 dev->dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size; 77 78 offset += dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size; 79 80 mqd_mem_obj->gtt_mem = (void *)((uint64_t)dev->dqm->hiq_sdma_mqd.gtt_mem 81 + offset); 82 mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr + offset; 83 mqd_mem_obj->cpu_ptr = (uint32_t *)((uint64_t) 84 dev->dqm->hiq_sdma_mqd.cpu_ptr + offset); 85 86 return mqd_mem_obj; 87 } 88 89 void free_mqd_hiq_sdma(struct mqd_manager *mm, void *mqd, 90 struct kfd_mem_obj *mqd_mem_obj) 91 { 92 WARN_ON(!mqd_mem_obj->gtt_mem); 93 kfree(mqd_mem_obj); 94 } 95 96 void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm, 97 const uint32_t *cu_mask, uint32_t cu_mask_count, 98 uint32_t *se_mask) 99 { 100 struct kfd_cu_info cu_info; 101 uint32_t cu_per_se[KFD_MAX_NUM_SE] = {0}; 102 int i, se, sh, cu = 0; 103 104 amdgpu_amdkfd_get_cu_info(mm->dev->kgd, &cu_info); 105 106 if (cu_mask_count > cu_info.cu_active_number) 107 cu_mask_count = cu_info.cu_active_number; 108 109 for (se = 0; se < cu_info.num_shader_engines; se++) 110 for (sh = 0; sh < cu_info.num_shader_arrays_per_engine; sh++) 111 cu_per_se[se] += hweight32(cu_info.cu_bitmap[se % 4][sh + (se / 4)]); 112 113 /* Symmetrically map cu_mask to all SEs: 114 * cu_mask[0] bit0 -> se_mask[0] bit0; 115 * cu_mask[0] bit1 -> se_mask[1] bit0; 116 * ... (if # SE is 4) 117 * cu_mask[0] bit4 -> se_mask[0] bit1; 118 * ... 119 */ 120 se = 0; 121 for (i = 0; i < cu_mask_count; i++) { 122 if (cu_mask[i / 32] & (1 << (i % 32))) 123 se_mask[se] |= 1 << cu; 124 125 do { 126 se++; 127 if (se == cu_info.num_shader_engines) { 128 se = 0; 129 cu++; 130 } 131 } while (cu >= cu_per_se[se] && cu < 32); 132 } 133 } 134