MAINTAINERS: Add phy-miphy28lp.c and phy-miphy365x.c to ARCH/STI architecture
[deliverable/linux.git] / drivers / gpu / drm / amd / amdkfd / kfd_kernel_queue.c
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 <linux/types.h>
25 #include <linux/mutex.h>
26 #include <linux/slab.h>
27 #include <linux/printk.h>
28 #include <linux/sched.h>
29 #include "kfd_kernel_queue.h"
30 #include "kfd_priv.h"
31 #include "kfd_device_queue_manager.h"
32 #include "kfd_pm4_headers.h"
33 #include "kfd_pm4_opcodes.h"
34
35 #define PM4_COUNT_ZERO (((1 << 15) - 1) << 16)
36
37 static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
38 enum kfd_queue_type type, unsigned int queue_size)
39 {
40 struct queue_properties prop;
41 int retval;
42 union PM4_MES_TYPE_3_HEADER nop;
43
44 BUG_ON(!kq || !dev);
45 BUG_ON(type != KFD_QUEUE_TYPE_DIQ && type != KFD_QUEUE_TYPE_HIQ);
46
47 pr_debug("kfd: In func %s initializing queue type %d size %d\n",
48 __func__, KFD_QUEUE_TYPE_HIQ, queue_size);
49
50 nop.opcode = IT_NOP;
51 nop.type = PM4_TYPE_3;
52 nop.u32all |= PM4_COUNT_ZERO;
53
54 kq->dev = dev;
55 kq->nop_packet = nop.u32all;
56 switch (type) {
57 case KFD_QUEUE_TYPE_DIQ:
58 case KFD_QUEUE_TYPE_HIQ:
59 kq->mqd = dev->dqm->ops.get_mqd_manager(dev->dqm,
60 KFD_MQD_TYPE_HIQ);
61 break;
62 default:
63 BUG();
64 break;
65 }
66
67 if (kq->mqd == NULL)
68 return false;
69
70 prop.doorbell_ptr = kfd_get_kernel_doorbell(dev, &prop.doorbell_off);
71
72 if (prop.doorbell_ptr == NULL)
73 goto err_get_kernel_doorbell;
74
75 retval = kfd_gtt_sa_allocate(dev, queue_size, &kq->pq);
76 if (retval != 0)
77 goto err_pq_allocate_vidmem;
78
79 kq->pq_kernel_addr = kq->pq->cpu_ptr;
80 kq->pq_gpu_addr = kq->pq->gpu_addr;
81
82 retval = kq->ops_asic_specific.initialize(kq, dev, type, queue_size);
83 if (retval == false)
84 goto err_eop_allocate_vidmem;
85
86 retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->rptr_kernel),
87 &kq->rptr_mem);
88
89 if (retval != 0)
90 goto err_rptr_allocate_vidmem;
91
92 kq->rptr_kernel = kq->rptr_mem->cpu_ptr;
93 kq->rptr_gpu_addr = kq->rptr_mem->gpu_addr;
94
95 retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->wptr_kernel),
96 &kq->wptr_mem);
97
98 if (retval != 0)
99 goto err_wptr_allocate_vidmem;
100
101 kq->wptr_kernel = kq->wptr_mem->cpu_ptr;
102 kq->wptr_gpu_addr = kq->wptr_mem->gpu_addr;
103
104 memset(kq->pq_kernel_addr, 0, queue_size);
105 memset(kq->rptr_kernel, 0, sizeof(*kq->rptr_kernel));
106 memset(kq->wptr_kernel, 0, sizeof(*kq->wptr_kernel));
107
108 prop.queue_size = queue_size;
109 prop.is_interop = false;
110 prop.priority = 1;
111 prop.queue_percent = 100;
112 prop.type = type;
113 prop.vmid = 0;
114 prop.queue_address = kq->pq_gpu_addr;
115 prop.read_ptr = (uint32_t *) kq->rptr_gpu_addr;
116 prop.write_ptr = (uint32_t *) kq->wptr_gpu_addr;
117 prop.eop_ring_buffer_address = kq->eop_gpu_addr;
118 prop.eop_ring_buffer_size = PAGE_SIZE;
119
120 if (init_queue(&kq->queue, prop) != 0)
121 goto err_init_queue;
122
123 kq->queue->device = dev;
124 kq->queue->process = kfd_get_process(current);
125
126 retval = kq->mqd->init_mqd(kq->mqd, &kq->queue->mqd,
127 &kq->queue->mqd_mem_obj,
128 &kq->queue->gart_mqd_addr,
129 &kq->queue->properties);
130 if (retval != 0)
131 goto err_init_mqd;
132
133 /* assign HIQ to HQD */
134 if (type == KFD_QUEUE_TYPE_HIQ) {
135 pr_debug("assigning hiq to hqd\n");
136 kq->queue->pipe = KFD_CIK_HIQ_PIPE;
137 kq->queue->queue = KFD_CIK_HIQ_QUEUE;
138 kq->mqd->load_mqd(kq->mqd, kq->queue->mqd, kq->queue->pipe,
139 kq->queue->queue, NULL);
140 } else {
141 /* allocate fence for DIQ */
142
143 retval = kfd_gtt_sa_allocate(dev, sizeof(uint32_t),
144 &kq->fence_mem_obj);
145
146 if (retval != 0)
147 goto err_alloc_fence;
148
149 kq->fence_kernel_address = kq->fence_mem_obj->cpu_ptr;
150 kq->fence_gpu_addr = kq->fence_mem_obj->gpu_addr;
151 }
152
153 print_queue(kq->queue);
154
155 return true;
156 err_alloc_fence:
157 err_init_mqd:
158 uninit_queue(kq->queue);
159 err_init_queue:
160 kfd_gtt_sa_free(dev, kq->wptr_mem);
161 err_wptr_allocate_vidmem:
162 kfd_gtt_sa_free(dev, kq->rptr_mem);
163 err_rptr_allocate_vidmem:
164 kfd_gtt_sa_free(dev, kq->eop_mem);
165 err_eop_allocate_vidmem:
166 kfd_gtt_sa_free(dev, kq->pq);
167 err_pq_allocate_vidmem:
168 pr_err("kfd: error init pq\n");
169 kfd_release_kernel_doorbell(dev, prop.doorbell_ptr);
170 err_get_kernel_doorbell:
171 pr_err("kfd: error init doorbell");
172 return false;
173
174 }
175
176 static void uninitialize(struct kernel_queue *kq)
177 {
178 BUG_ON(!kq);
179
180 if (kq->queue->properties.type == KFD_QUEUE_TYPE_HIQ)
181 kq->mqd->destroy_mqd(kq->mqd,
182 NULL,
183 false,
184 QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS,
185 kq->queue->pipe,
186 kq->queue->queue);
187 else if (kq->queue->properties.type == KFD_QUEUE_TYPE_DIQ)
188 kfd_gtt_sa_free(kq->dev, kq->fence_mem_obj);
189
190 kfd_gtt_sa_free(kq->dev, kq->rptr_mem);
191 kfd_gtt_sa_free(kq->dev, kq->wptr_mem);
192 kq->ops_asic_specific.uninitialize(kq);
193 kfd_gtt_sa_free(kq->dev, kq->pq);
194 kfd_release_kernel_doorbell(kq->dev,
195 kq->queue->properties.doorbell_ptr);
196 uninit_queue(kq->queue);
197 }
198
199 static int acquire_packet_buffer(struct kernel_queue *kq,
200 size_t packet_size_in_dwords, unsigned int **buffer_ptr)
201 {
202 size_t available_size;
203 size_t queue_size_dwords;
204 uint32_t wptr, rptr;
205 unsigned int *queue_address;
206
207 BUG_ON(!kq || !buffer_ptr);
208
209 rptr = *kq->rptr_kernel;
210 wptr = *kq->wptr_kernel;
211 queue_address = (unsigned int *)kq->pq_kernel_addr;
212 queue_size_dwords = kq->queue->properties.queue_size / sizeof(uint32_t);
213
214 pr_debug("kfd: In func %s\nrptr: %d\nwptr: %d\nqueue_address 0x%p\n",
215 __func__, rptr, wptr, queue_address);
216
217 available_size = (rptr - 1 - wptr + queue_size_dwords) %
218 queue_size_dwords;
219
220 if (packet_size_in_dwords >= queue_size_dwords ||
221 packet_size_in_dwords >= available_size) {
222 /*
223 * make sure calling functions know
224 * acquire_packet_buffer() failed
225 */
226 *buffer_ptr = NULL;
227 return -ENOMEM;
228 }
229
230 if (wptr + packet_size_in_dwords >= queue_size_dwords) {
231 while (wptr > 0) {
232 queue_address[wptr] = kq->nop_packet;
233 wptr = (wptr + 1) % queue_size_dwords;
234 }
235 }
236
237 *buffer_ptr = &queue_address[wptr];
238 kq->pending_wptr = wptr + packet_size_in_dwords;
239
240 return 0;
241 }
242
243 static void submit_packet(struct kernel_queue *kq)
244 {
245 #ifdef DEBUG
246 int i;
247 #endif
248
249 BUG_ON(!kq);
250
251 #ifdef DEBUG
252 for (i = *kq->wptr_kernel; i < kq->pending_wptr; i++) {
253 pr_debug("0x%2X ", kq->pq_kernel_addr[i]);
254 if (i % 15 == 0)
255 pr_debug("\n");
256 }
257 pr_debug("\n");
258 #endif
259
260 *kq->wptr_kernel = kq->pending_wptr;
261 write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
262 kq->pending_wptr);
263 }
264
265 static void rollback_packet(struct kernel_queue *kq)
266 {
267 BUG_ON(!kq);
268 kq->pending_wptr = *kq->queue->properties.write_ptr;
269 }
270
271 struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
272 enum kfd_queue_type type)
273 {
274 struct kernel_queue *kq;
275
276 BUG_ON(!dev);
277
278 kq = kzalloc(sizeof(struct kernel_queue), GFP_KERNEL);
279 if (!kq)
280 return NULL;
281
282 kq->ops.initialize = initialize;
283 kq->ops.uninitialize = uninitialize;
284 kq->ops.acquire_packet_buffer = acquire_packet_buffer;
285 kq->ops.submit_packet = submit_packet;
286 kq->ops.rollback_packet = rollback_packet;
287
288 switch (dev->device_info->asic_family) {
289 case CHIP_CARRIZO:
290 kernel_queue_init_vi(&kq->ops_asic_specific);
291 break;
292
293 case CHIP_KAVERI:
294 kernel_queue_init_cik(&kq->ops_asic_specific);
295 break;
296 }
297
298 if (kq->ops.initialize(kq, dev, type, KFD_KERNEL_QUEUE_SIZE) == false) {
299 pr_err("kfd: failed to init kernel queue\n");
300 kfree(kq);
301 return NULL;
302 }
303 return kq;
304 }
305
306 void kernel_queue_uninit(struct kernel_queue *kq)
307 {
308 BUG_ON(!kq);
309
310 kq->ops.uninitialize(kq);
311 kfree(kq);
312 }
313
314 static __attribute__((unused)) void test_kq(struct kfd_dev *dev)
315 {
316 struct kernel_queue *kq;
317 uint32_t *buffer, i;
318 int retval;
319
320 BUG_ON(!dev);
321
322 pr_err("kfd: starting kernel queue test\n");
323
324 kq = kernel_queue_init(dev, KFD_QUEUE_TYPE_HIQ);
325 BUG_ON(!kq);
326
327 retval = kq->ops.acquire_packet_buffer(kq, 5, &buffer);
328 BUG_ON(retval != 0);
329 for (i = 0; i < 5; i++)
330 buffer[i] = kq->nop_packet;
331 kq->ops.submit_packet(kq);
332
333 pr_err("kfd: ending kernel queue test\n");
334 }
335
336
This page took 0.084249 seconds and 5 git commands to generate.