drm/amdkfd: Add gtt sa related data to kfd_dev struct
[deliverable/linux.git] / drivers / gpu / drm / amd / amdkfd / kfd_device.c
CommitLineData
4a488a7a
OG
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#include <linux/amd-iommu.h>
24#include <linux/bsearch.h>
25#include <linux/pci.h>
26#include <linux/slab.h>
27#include "kfd_priv.h"
64c7f8cf 28#include "kfd_device_queue_manager.h"
4a488a7a 29
19f6d2a6
OG
30#define MQD_SIZE_ALIGNED 768
31
4a488a7a
OG
32static const struct kfd_device_info kaveri_device_info = {
33 .max_pasid_bits = 16,
b3f5e6b4 34 .ih_ring_entry_size = 4 * sizeof(uint32_t),
f7c826ad 35 .num_of_watch_points = 4,
19f6d2a6 36 .mqd_size_aligned = MQD_SIZE_ALIGNED
4a488a7a
OG
37};
38
39struct kfd_deviceid {
40 unsigned short did;
41 const struct kfd_device_info *device_info;
42};
43
44/* Please keep this sorted by increasing device id. */
45static const struct kfd_deviceid supported_devices[] = {
46 { 0x1304, &kaveri_device_info }, /* Kaveri */
47 { 0x1305, &kaveri_device_info }, /* Kaveri */
48 { 0x1306, &kaveri_device_info }, /* Kaveri */
49 { 0x1307, &kaveri_device_info }, /* Kaveri */
50 { 0x1309, &kaveri_device_info }, /* Kaveri */
51 { 0x130A, &kaveri_device_info }, /* Kaveri */
52 { 0x130B, &kaveri_device_info }, /* Kaveri */
53 { 0x130C, &kaveri_device_info }, /* Kaveri */
54 { 0x130D, &kaveri_device_info }, /* Kaveri */
55 { 0x130E, &kaveri_device_info }, /* Kaveri */
56 { 0x130F, &kaveri_device_info }, /* Kaveri */
57 { 0x1310, &kaveri_device_info }, /* Kaveri */
58 { 0x1311, &kaveri_device_info }, /* Kaveri */
59 { 0x1312, &kaveri_device_info }, /* Kaveri */
60 { 0x1313, &kaveri_device_info }, /* Kaveri */
61 { 0x1315, &kaveri_device_info }, /* Kaveri */
62 { 0x1316, &kaveri_device_info }, /* Kaveri */
63 { 0x1317, &kaveri_device_info }, /* Kaveri */
64 { 0x1318, &kaveri_device_info }, /* Kaveri */
65 { 0x131B, &kaveri_device_info }, /* Kaveri */
66 { 0x131C, &kaveri_device_info }, /* Kaveri */
67 { 0x131D, &kaveri_device_info }, /* Kaveri */
68};
69
70static const struct kfd_device_info *lookup_device_info(unsigned short did)
71{
72 size_t i;
73
74 for (i = 0; i < ARRAY_SIZE(supported_devices); i++) {
75 if (supported_devices[i].did == did) {
76 BUG_ON(supported_devices[i].device_info == NULL);
77 return supported_devices[i].device_info;
78 }
79 }
80
81 return NULL;
82}
83
84struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev)
85{
86 struct kfd_dev *kfd;
87
88 const struct kfd_device_info *device_info =
89 lookup_device_info(pdev->device);
90
91 if (!device_info)
92 return NULL;
93
94 kfd = kzalloc(sizeof(*kfd), GFP_KERNEL);
95 if (!kfd)
96 return NULL;
97
98 kfd->kgd = kgd;
99 kfd->device_info = device_info;
100 kfd->pdev = pdev;
19f6d2a6 101 kfd->init_complete = false;
4a488a7a
OG
102
103 return kfd;
104}
105
b17f068a
OG
106static bool device_iommu_pasid_init(struct kfd_dev *kfd)
107{
108 const u32 required_iommu_flags = AMD_IOMMU_DEVICE_FLAG_ATS_SUP |
109 AMD_IOMMU_DEVICE_FLAG_PRI_SUP |
110 AMD_IOMMU_DEVICE_FLAG_PASID_SUP;
111
112 struct amd_iommu_device_info iommu_info;
113 unsigned int pasid_limit;
114 int err;
115
116 err = amd_iommu_device_info(kfd->pdev, &iommu_info);
117 if (err < 0) {
118 dev_err(kfd_device,
119 "error getting iommu info. is the iommu enabled?\n");
120 return false;
121 }
122
123 if ((iommu_info.flags & required_iommu_flags) != required_iommu_flags) {
124 dev_err(kfd_device, "error required iommu flags ats(%i), pri(%i), pasid(%i)\n",
125 (iommu_info.flags & AMD_IOMMU_DEVICE_FLAG_ATS_SUP) != 0,
126 (iommu_info.flags & AMD_IOMMU_DEVICE_FLAG_PRI_SUP) != 0,
127 (iommu_info.flags & AMD_IOMMU_DEVICE_FLAG_PASID_SUP) != 0);
128 return false;
129 }
130
131 pasid_limit = min_t(unsigned int,
132 (unsigned int)1 << kfd->device_info->max_pasid_bits,
133 iommu_info.max_pasids);
134 /*
135 * last pasid is used for kernel queues doorbells
136 * in the future the last pasid might be used for a kernel thread.
137 */
138 pasid_limit = min_t(unsigned int,
139 pasid_limit,
140 kfd->doorbell_process_limit - 1);
141
142 err = amd_iommu_init_device(kfd->pdev, pasid_limit);
143 if (err < 0) {
144 dev_err(kfd_device, "error initializing iommu device\n");
145 return false;
146 }
147
148 if (!kfd_set_pasid_limit(pasid_limit)) {
149 dev_err(kfd_device, "error setting pasid limit\n");
150 amd_iommu_free_device(kfd->pdev);
151 return false;
152 }
153
154 return true;
155}
156
157static void iommu_pasid_shutdown_callback(struct pci_dev *pdev, int pasid)
158{
159 struct kfd_dev *dev = kfd_device_by_pci_dev(pdev);
160
161 if (dev)
162 kfd_unbind_process_from_device(dev, pasid);
163}
164
4a488a7a
OG
165bool kgd2kfd_device_init(struct kfd_dev *kfd,
166 const struct kgd2kfd_shared_resources *gpu_resources)
167{
19f6d2a6
OG
168 unsigned int size;
169
4a488a7a
OG
170 kfd->shared_resources = *gpu_resources;
171
19f6d2a6
OG
172 /* calculate max size of mqds needed for queues */
173 size = max_num_of_processes *
174 max_num_of_queues_per_process *
175 kfd->device_info->mqd_size_aligned;
176
177 /* add another 512KB for all other allocations on gart */
178 size += 512 * 1024;
179
180 if (kfd2kgd->init_sa_manager(kfd->kgd, size)) {
181 dev_err(kfd_device,
182 "Error initializing sa manager for device (%x:%x)\n",
183 kfd->pdev->vendor, kfd->pdev->device);
184 goto out;
185 }
186
187 kfd_doorbell_init(kfd);
188
189 if (kfd_topology_add_device(kfd) != 0) {
190 dev_err(kfd_device,
191 "Error adding device (%x:%x) to topology\n",
192 kfd->pdev->vendor, kfd->pdev->device);
193 goto kfd_topology_add_device_error;
194 }
195
b3f5e6b4
AL
196 if (kfd_interrupt_init(kfd)) {
197 dev_err(kfd_device,
198 "Error initializing interrupts for device (%x:%x)\n",
199 kfd->pdev->vendor, kfd->pdev->device);
200 goto kfd_interrupt_error;
201 }
202
b17f068a
OG
203 if (!device_iommu_pasid_init(kfd)) {
204 dev_err(kfd_device,
205 "Error initializing iommuv2 for device (%x:%x)\n",
206 kfd->pdev->vendor, kfd->pdev->device);
207 goto device_iommu_pasid_error;
208 }
209 amd_iommu_set_invalidate_ctx_cb(kfd->pdev,
210 iommu_pasid_shutdown_callback);
5b5c4e40 211
64c7f8cf
BG
212 kfd->dqm = device_queue_manager_init(kfd);
213 if (!kfd->dqm) {
214 dev_err(kfd_device,
215 "Error initializing queue manager for device (%x:%x)\n",
216 kfd->pdev->vendor, kfd->pdev->device);
217 goto device_queue_manager_error;
218 }
219
220 if (kfd->dqm->start(kfd->dqm) != 0) {
221 dev_err(kfd_device,
222 "Error starting queuen manager for device (%x:%x)\n",
223 kfd->pdev->vendor, kfd->pdev->device);
224 goto dqm_start_error;
225 }
226
4a488a7a
OG
227 kfd->init_complete = true;
228 dev_info(kfd_device, "added device (%x:%x)\n", kfd->pdev->vendor,
229 kfd->pdev->device);
230
64c7f8cf
BG
231 pr_debug("kfd: Starting kfd with the following scheduling policy %d\n",
232 sched_policy);
233
19f6d2a6
OG
234 goto out;
235
64c7f8cf
BG
236dqm_start_error:
237 device_queue_manager_uninit(kfd->dqm);
238device_queue_manager_error:
239 amd_iommu_free_device(kfd->pdev);
b17f068a 240device_iommu_pasid_error:
b3f5e6b4
AL
241 kfd_interrupt_exit(kfd);
242kfd_interrupt_error:
b17f068a 243 kfd_topology_remove_device(kfd);
19f6d2a6
OG
244kfd_topology_add_device_error:
245 kfd2kgd->fini_sa_manager(kfd->kgd);
246 dev_err(kfd_device,
247 "device (%x:%x) NOT added due to errors\n",
248 kfd->pdev->vendor, kfd->pdev->device);
249out:
250 return kfd->init_complete;
4a488a7a
OG
251}
252
253void kgd2kfd_device_exit(struct kfd_dev *kfd)
254{
b17f068a 255 if (kfd->init_complete) {
64c7f8cf 256 device_queue_manager_uninit(kfd->dqm);
b17f068a 257 amd_iommu_free_device(kfd->pdev);
b3f5e6b4 258 kfd_interrupt_exit(kfd);
b17f068a
OG
259 kfd_topology_remove_device(kfd);
260 }
5b5c4e40 261
4a488a7a
OG
262 kfree(kfd);
263}
264
265void kgd2kfd_suspend(struct kfd_dev *kfd)
266{
267 BUG_ON(kfd == NULL);
b17f068a 268
64c7f8cf
BG
269 if (kfd->init_complete) {
270 kfd->dqm->stop(kfd->dqm);
abc9d3e3 271 amd_iommu_set_invalidate_ctx_cb(kfd->pdev, NULL);
b17f068a 272 amd_iommu_free_device(kfd->pdev);
64c7f8cf 273 }
4a488a7a
OG
274}
275
276int kgd2kfd_resume(struct kfd_dev *kfd)
277{
b17f068a
OG
278 unsigned int pasid_limit;
279 int err;
280
4a488a7a
OG
281 BUG_ON(kfd == NULL);
282
b17f068a
OG
283 pasid_limit = kfd_get_pasid_limit();
284
285 if (kfd->init_complete) {
286 err = amd_iommu_init_device(kfd->pdev, pasid_limit);
287 if (err < 0)
288 return -ENXIO;
289 amd_iommu_set_invalidate_ctx_cb(kfd->pdev,
290 iommu_pasid_shutdown_callback);
64c7f8cf 291 kfd->dqm->start(kfd->dqm);
b17f068a
OG
292 }
293
4a488a7a
OG
294 return 0;
295}
296
b3f5e6b4
AL
297/* This is called directly from KGD at ISR. */
298void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
4a488a7a 299{
b3f5e6b4
AL
300 if (kfd->init_complete) {
301 spin_lock(&kfd->interrupt_lock);
302
303 if (kfd->interrupts_active
304 && enqueue_ih_ring_entry(kfd, ih_ring_entry))
305 schedule_work(&kfd->interrupt_work);
306
307 spin_unlock(&kfd->interrupt_lock);
308 }
4a488a7a 309}
This page took 0.051821 seconds and 5 git commands to generate.