1 /*
2  * Copyright(c) 2011-2016 Intel Corporation. All rights reserved.
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 (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * 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 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  *
23  * Authors:
24  *    Kevin Tian <kevin.tian@intel.com>
25  *    Dexuan Cui
26  *
27  * Contributors:
28  *    Pei Zhang <pei.zhang@intel.com>
29  *    Min He <min.he@intel.com>
30  *    Niu Bing <bing.niu@intel.com>
31  *    Yulei Zhang <yulei.zhang@intel.com>
32  *    Zhenyu Wang <zhenyuw@linux.intel.com>
33  *    Zhi Wang <zhi.a.wang@intel.com>
34  *
35  */
36 
37 #include "i915_drv.h"
38 #include "gvt.h"
39 
40 #define MB_TO_BYTES(mb) ((mb) << 20ULL)
41 #define BYTES_TO_MB(b) ((b) >> 20ULL)
42 
43 #define HOST_LOW_GM_SIZE MB_TO_BYTES(128)
44 #define HOST_HIGH_GM_SIZE MB_TO_BYTES(384)
45 #define HOST_FENCE 4
46 
47 static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
48 {
49 	struct intel_gvt *gvt = vgpu->gvt;
50 	struct drm_i915_private *dev_priv = gvt->dev_priv;
51 	u32 alloc_flag, search_flag;
52 	u64 start, end, size;
53 	struct drm_mm_node *node;
54 	int retried = 0;
55 	int ret;
56 
57 	if (high_gm) {
58 		search_flag = DRM_MM_SEARCH_BELOW;
59 		alloc_flag = DRM_MM_CREATE_TOP;
60 		node = &vgpu->gm.high_gm_node;
61 		size = vgpu_hidden_sz(vgpu);
62 		start = gvt_hidden_gmadr_base(gvt);
63 		end = gvt_hidden_gmadr_end(gvt);
64 	} else {
65 		search_flag = DRM_MM_SEARCH_DEFAULT;
66 		alloc_flag = DRM_MM_CREATE_DEFAULT;
67 		node = &vgpu->gm.low_gm_node;
68 		size = vgpu_aperture_sz(vgpu);
69 		start = gvt_aperture_gmadr_base(gvt);
70 		end = gvt_aperture_gmadr_end(gvt);
71 	}
72 
73 	mutex_lock(&dev_priv->drm.struct_mutex);
74 search_again:
75 	ret = drm_mm_insert_node_in_range_generic(&dev_priv->ggtt.base.mm,
76 						  node, size, 4096, 0,
77 						  start, end, search_flag,
78 						  alloc_flag);
79 	if (ret) {
80 		ret = i915_gem_evict_something(&dev_priv->ggtt.base,
81 					       size, 4096, 0, start, end, 0);
82 		if (ret == 0 && ++retried < 3)
83 			goto search_again;
84 
85 		gvt_err("fail to alloc %s gm space from host, retried %d\n",
86 				high_gm ? "high" : "low", retried);
87 	}
88 	mutex_unlock(&dev_priv->drm.struct_mutex);
89 	return ret;
90 }
91 
92 static int alloc_vgpu_gm(struct intel_vgpu *vgpu)
93 {
94 	struct intel_gvt *gvt = vgpu->gvt;
95 	struct drm_i915_private *dev_priv = gvt->dev_priv;
96 	int ret;
97 
98 	ret = alloc_gm(vgpu, false);
99 	if (ret)
100 		return ret;
101 
102 	ret = alloc_gm(vgpu, true);
103 	if (ret)
104 		goto out_free_aperture;
105 
106 	gvt_dbg_core("vgpu%d: alloc low GM start %llx size %llx\n", vgpu->id,
107 		     vgpu_aperture_offset(vgpu), vgpu_aperture_sz(vgpu));
108 
109 	gvt_dbg_core("vgpu%d: alloc high GM start %llx size %llx\n", vgpu->id,
110 		     vgpu_hidden_offset(vgpu), vgpu_hidden_sz(vgpu));
111 
112 	return 0;
113 out_free_aperture:
114 	mutex_lock(&dev_priv->drm.struct_mutex);
115 	drm_mm_remove_node(&vgpu->gm.low_gm_node);
116 	mutex_unlock(&dev_priv->drm.struct_mutex);
117 	return ret;
118 }
119 
120 static void free_vgpu_gm(struct intel_vgpu *vgpu)
121 {
122 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
123 
124 	mutex_lock(&dev_priv->drm.struct_mutex);
125 	drm_mm_remove_node(&vgpu->gm.low_gm_node);
126 	drm_mm_remove_node(&vgpu->gm.high_gm_node);
127 	mutex_unlock(&dev_priv->drm.struct_mutex);
128 }
129 
130 /**
131  * intel_vgpu_write_fence - write fence registers owned by a vGPU
132  * @vgpu: vGPU instance
133  * @fence: vGPU fence register number
134  * @value: Fence register value to be written
135  *
136  * This function is used to write fence registers owned by a vGPU. The vGPU
137  * fence register number will be translated into HW fence register number.
138  *
139  */
140 void intel_vgpu_write_fence(struct intel_vgpu *vgpu,
141 		u32 fence, u64 value)
142 {
143 	struct intel_gvt *gvt = vgpu->gvt;
144 	struct drm_i915_private *dev_priv = gvt->dev_priv;
145 	struct drm_i915_fence_reg *reg;
146 	i915_reg_t fence_reg_lo, fence_reg_hi;
147 
148 	assert_rpm_wakelock_held(dev_priv);
149 
150 	if (WARN_ON(fence > vgpu_fence_sz(vgpu)))
151 		return;
152 
153 	reg = vgpu->fence.regs[fence];
154 	if (WARN_ON(!reg))
155 		return;
156 
157 	fence_reg_lo = FENCE_REG_GEN6_LO(reg->id);
158 	fence_reg_hi = FENCE_REG_GEN6_HI(reg->id);
159 
160 	I915_WRITE(fence_reg_lo, 0);
161 	POSTING_READ(fence_reg_lo);
162 
163 	I915_WRITE(fence_reg_hi, upper_32_bits(value));
164 	I915_WRITE(fence_reg_lo, lower_32_bits(value));
165 	POSTING_READ(fence_reg_lo);
166 }
167 
168 static void free_vgpu_fence(struct intel_vgpu *vgpu)
169 {
170 	struct intel_gvt *gvt = vgpu->gvt;
171 	struct drm_i915_private *dev_priv = gvt->dev_priv;
172 	struct drm_i915_fence_reg *reg;
173 	u32 i;
174 
175 	if (WARN_ON(!vgpu_fence_sz(vgpu)))
176 		return;
177 
178 	intel_runtime_pm_get(dev_priv);
179 
180 	mutex_lock(&dev_priv->drm.struct_mutex);
181 	for (i = 0; i < vgpu_fence_sz(vgpu); i++) {
182 		reg = vgpu->fence.regs[i];
183 		intel_vgpu_write_fence(vgpu, i, 0);
184 		list_add_tail(&reg->link,
185 			      &dev_priv->mm.fence_list);
186 	}
187 	mutex_unlock(&dev_priv->drm.struct_mutex);
188 
189 	intel_runtime_pm_put(dev_priv);
190 }
191 
192 static int alloc_vgpu_fence(struct intel_vgpu *vgpu)
193 {
194 	struct intel_gvt *gvt = vgpu->gvt;
195 	struct drm_i915_private *dev_priv = gvt->dev_priv;
196 	struct drm_i915_fence_reg *reg;
197 	int i;
198 	struct list_head *pos, *q;
199 
200 	intel_runtime_pm_get(dev_priv);
201 
202 	/* Request fences from host */
203 	mutex_lock(&dev_priv->drm.struct_mutex);
204 	i = 0;
205 	list_for_each_safe(pos, q, &dev_priv->mm.fence_list) {
206 		reg = list_entry(pos, struct drm_i915_fence_reg, link);
207 		if (reg->pin_count || reg->vma)
208 			continue;
209 		list_del(pos);
210 		vgpu->fence.regs[i] = reg;
211 		intel_vgpu_write_fence(vgpu, i, 0);
212 		if (++i == vgpu_fence_sz(vgpu))
213 			break;
214 	}
215 	if (i != vgpu_fence_sz(vgpu))
216 		goto out_free_fence;
217 
218 	mutex_unlock(&dev_priv->drm.struct_mutex);
219 	intel_runtime_pm_put(dev_priv);
220 	return 0;
221 out_free_fence:
222 	/* Return fences to host, if fail */
223 	for (i = 0; i < vgpu_fence_sz(vgpu); i++) {
224 		reg = vgpu->fence.regs[i];
225 		if (!reg)
226 			continue;
227 		list_add_tail(&reg->link,
228 			      &dev_priv->mm.fence_list);
229 	}
230 	mutex_unlock(&dev_priv->drm.struct_mutex);
231 	intel_runtime_pm_put(dev_priv);
232 	return -ENOSPC;
233 }
234 
235 static void free_resource(struct intel_vgpu *vgpu)
236 {
237 	struct intel_gvt *gvt = vgpu->gvt;
238 
239 	gvt->gm.vgpu_allocated_low_gm_size -= vgpu_aperture_sz(vgpu);
240 	gvt->gm.vgpu_allocated_high_gm_size -= vgpu_hidden_sz(vgpu);
241 	gvt->fence.vgpu_allocated_fence_num -= vgpu_fence_sz(vgpu);
242 }
243 
244 static int alloc_resource(struct intel_vgpu *vgpu,
245 		struct intel_vgpu_creation_params *param)
246 {
247 	struct intel_gvt *gvt = vgpu->gvt;
248 	unsigned long request, avail, max, taken;
249 	const char *item;
250 
251 	if (!param->low_gm_sz || !param->high_gm_sz || !param->fence_sz) {
252 		gvt_err("Invalid vGPU creation params\n");
253 		return -EINVAL;
254 	}
255 
256 	item = "low GM space";
257 	max = gvt_aperture_sz(gvt) - HOST_LOW_GM_SIZE;
258 	taken = gvt->gm.vgpu_allocated_low_gm_size;
259 	avail = max - taken;
260 	request = MB_TO_BYTES(param->low_gm_sz);
261 
262 	if (request > avail)
263 		goto no_enough_resource;
264 
265 	vgpu_aperture_sz(vgpu) = request;
266 
267 	item = "high GM space";
268 	max = gvt_hidden_sz(gvt) - HOST_HIGH_GM_SIZE;
269 	taken = gvt->gm.vgpu_allocated_high_gm_size;
270 	avail = max - taken;
271 	request = MB_TO_BYTES(param->high_gm_sz);
272 
273 	if (request > avail)
274 		goto no_enough_resource;
275 
276 	vgpu_hidden_sz(vgpu) = request;
277 
278 	item = "fence";
279 	max = gvt_fence_sz(gvt) - HOST_FENCE;
280 	taken = gvt->fence.vgpu_allocated_fence_num;
281 	avail = max - taken;
282 	request = param->fence_sz;
283 
284 	if (request > avail)
285 		goto no_enough_resource;
286 
287 	vgpu_fence_sz(vgpu) = request;
288 
289 	gvt->gm.vgpu_allocated_low_gm_size += MB_TO_BYTES(param->low_gm_sz);
290 	gvt->gm.vgpu_allocated_high_gm_size += MB_TO_BYTES(param->high_gm_sz);
291 	gvt->fence.vgpu_allocated_fence_num += param->fence_sz;
292 	return 0;
293 
294 no_enough_resource:
295 	gvt_err("vgpu%d: fail to allocate resource %s\n", vgpu->id, item);
296 	gvt_err("vgpu%d: request %luMB avail %luMB max %luMB taken %luMB\n",
297 		vgpu->id, BYTES_TO_MB(request), BYTES_TO_MB(avail),
298 		BYTES_TO_MB(max), BYTES_TO_MB(taken));
299 	return -ENOSPC;
300 }
301 
302 /**
303  * inte_gvt_free_vgpu_resource - free HW resource owned by a vGPU
304  * @vgpu: a vGPU
305  *
306  * This function is used to free the HW resource owned by a vGPU.
307  *
308  */
309 void intel_vgpu_free_resource(struct intel_vgpu *vgpu)
310 {
311 	free_vgpu_gm(vgpu);
312 	free_vgpu_fence(vgpu);
313 	free_resource(vgpu);
314 }
315 
316 /**
317  * intel_alloc_vgpu_resource - allocate HW resource for a vGPU
318  * @vgpu: vGPU
319  * @param: vGPU creation params
320  *
321  * This function is used to allocate HW resource for a vGPU. User specifies
322  * the resource configuration through the creation params.
323  *
324  * Returns:
325  * zero on success, negative error code if failed.
326  *
327  */
328 int intel_vgpu_alloc_resource(struct intel_vgpu *vgpu,
329 		struct intel_vgpu_creation_params *param)
330 {
331 	int ret;
332 
333 	ret = alloc_resource(vgpu, param);
334 	if (ret)
335 		return ret;
336 
337 	ret = alloc_vgpu_gm(vgpu);
338 	if (ret)
339 		goto out_free_resource;
340 
341 	ret = alloc_vgpu_fence(vgpu);
342 	if (ret)
343 		goto out_free_vgpu_gm;
344 
345 	return 0;
346 
347 out_free_vgpu_gm:
348 	free_vgpu_gm(vgpu);
349 out_free_resource:
350 	free_resource(vgpu);
351 	return ret;
352 }
353