xref: /openbmc/linux/drivers/gpu/drm/i915/gvt/mmio_context.c (revision 05cf4fe738242183f1237f1b3a28b4479348c0a1)
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  *    Eddie Dong <eddie.dong@intel.com>
25  *    Kevin Tian <kevin.tian@intel.com>
26  *
27  * Contributors:
28  *    Zhi Wang <zhi.a.wang@intel.com>
29  *    Changbin Du <changbin.du@intel.com>
30  *    Zhenyu Wang <zhenyuw@linux.intel.com>
31  *    Tina Zhang <tina.zhang@intel.com>
32  *    Bing Niu <bing.niu@intel.com>
33  *
34  */
35 
36 #include "i915_drv.h"
37 #include "gvt.h"
38 #include "trace.h"
39 
40 #define GEN9_MOCS_SIZE		64
41 
42 /* Raw offset is appened to each line for convenience. */
43 static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = {
44 	{RCS, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
45 	{RCS, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
46 	{RCS, HWSTAM, 0x0, false}, /* 0x2098 */
47 	{RCS, INSTPM, 0xffff, true}, /* 0x20c0 */
48 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
49 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
50 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
51 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
52 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
53 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
54 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
55 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
56 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
57 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
58 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
59 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
60 	{RCS, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
61 	{RCS, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
62 	{RCS, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
63 	{RCS, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
64 	{RCS, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
65 	{RCS, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
66 
67 	{BCS, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
68 	{BCS, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
69 	{BCS, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
70 	{BCS, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
71 	{BCS, RING_EXCC(BLT_RING_BASE), 0x0, false}, /* 0x22028 */
72 	{RCS, INVALID_MMIO_REG, 0, false } /* Terminated */
73 };
74 
75 static struct engine_mmio gen9_engine_mmio_list[] __cacheline_aligned = {
76 	{RCS, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
77 	{RCS, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
78 	{RCS, HWSTAM, 0x0, false}, /* 0x2098 */
79 	{RCS, INSTPM, 0xffff, true}, /* 0x20c0 */
80 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
81 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
82 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
83 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
84 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
85 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
86 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
87 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
88 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
89 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
90 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
91 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
92 	{RCS, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
93 	{RCS, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
94 	{RCS, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
95 	{RCS, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
96 	{RCS, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
97 	{RCS, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
98 
99 	{RCS, GEN8_PRIVATE_PAT_LO, 0, false}, /* 0x40e0 */
100 	{RCS, GEN8_PRIVATE_PAT_HI, 0, false}, /* 0x40e4 */
101 	{RCS, GEN8_CS_CHICKEN1, 0xffff, true}, /* 0x2580 */
102 	{RCS, COMMON_SLICE_CHICKEN2, 0xffff, true}, /* 0x7014 */
103 	{RCS, GEN9_CS_DEBUG_MODE1, 0xffff, false}, /* 0x20ec */
104 	{RCS, GEN8_L3SQCREG4, 0, false}, /* 0xb118 */
105 	{RCS, GEN7_HALF_SLICE_CHICKEN1, 0xffff, true}, /* 0xe100 */
106 	{RCS, HALF_SLICE_CHICKEN2, 0xffff, true}, /* 0xe180 */
107 	{RCS, HALF_SLICE_CHICKEN3, 0xffff, true}, /* 0xe184 */
108 	{RCS, GEN9_HALF_SLICE_CHICKEN5, 0xffff, true}, /* 0xe188 */
109 	{RCS, GEN9_HALF_SLICE_CHICKEN7, 0xffff, true}, /* 0xe194 */
110 	{RCS, GEN8_ROW_CHICKEN, 0xffff, true}, /* 0xe4f0 */
111 	{RCS, TRVATTL3PTRDW(0), 0, false}, /* 0x4de0 */
112 	{RCS, TRVATTL3PTRDW(1), 0, false}, /* 0x4de4 */
113 	{RCS, TRNULLDETCT, 0, false}, /* 0x4de8 */
114 	{RCS, TRINVTILEDETCT, 0, false}, /* 0x4dec */
115 	{RCS, TRVADR, 0, false}, /* 0x4df0 */
116 	{RCS, TRTTE, 0, false}, /* 0x4df4 */
117 
118 	{BCS, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
119 	{BCS, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
120 	{BCS, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
121 	{BCS, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
122 	{BCS, RING_EXCC(BLT_RING_BASE), 0x0, false}, /* 0x22028 */
123 
124 	{VCS2, RING_EXCC(GEN8_BSD2_RING_BASE), 0xffff, false}, /* 0x1c028 */
125 
126 	{VECS, RING_EXCC(VEBOX_RING_BASE), 0xffff, false}, /* 0x1a028 */
127 
128 	{RCS, GEN8_HDC_CHICKEN1, 0xffff, true}, /* 0x7304 */
129 	{RCS, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
130 	{RCS, GEN7_UCGCTL4, 0x0, false}, /* 0x940c */
131 	{RCS, GAMT_CHKN_BIT_REG, 0x0, false}, /* 0x4ab8 */
132 
133 	{RCS, GEN9_GAMT_ECO_REG_RW_IA, 0x0, false}, /* 0x4ab0 */
134 	{RCS, GEN9_CSFE_CHICKEN1_RCS, 0xffff, false}, /* 0x20d4 */
135 
136 	{RCS, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */
137 	{RCS, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */
138 	{RCS, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */
139 	{RCS, INVALID_MMIO_REG, 0, false } /* Terminated */
140 };
141 
142 static struct {
143 	bool initialized;
144 	u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
145 	u32 l3cc_table[GEN9_MOCS_SIZE / 2];
146 } gen9_render_mocs;
147 
148 static void load_render_mocs(struct drm_i915_private *dev_priv)
149 {
150 	i915_reg_t offset;
151 	u32 regs[] = {
152 		[RCS] = 0xc800,
153 		[VCS] = 0xc900,
154 		[VCS2] = 0xca00,
155 		[BCS] = 0xcc00,
156 		[VECS] = 0xcb00,
157 	};
158 	int ring_id, i;
159 
160 	for (ring_id = 0; ring_id < ARRAY_SIZE(regs); ring_id++) {
161 		if (!HAS_ENGINE(dev_priv, ring_id))
162 			continue;
163 		offset.reg = regs[ring_id];
164 		for (i = 0; i < GEN9_MOCS_SIZE; i++) {
165 			gen9_render_mocs.control_table[ring_id][i] =
166 				I915_READ_FW(offset);
167 			offset.reg += 4;
168 		}
169 	}
170 
171 	offset.reg = 0xb020;
172 	for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
173 		gen9_render_mocs.l3cc_table[i] =
174 			I915_READ_FW(offset);
175 		offset.reg += 4;
176 	}
177 	gen9_render_mocs.initialized = true;
178 }
179 
180 static int
181 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
182 				 struct i915_request *req)
183 {
184 	u32 *cs;
185 	int ret;
186 	struct engine_mmio *mmio;
187 	struct intel_gvt *gvt = vgpu->gvt;
188 	int ring_id = req->engine->id;
189 	int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
190 
191 	if (count == 0)
192 		return 0;
193 
194 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
195 	if (ret)
196 		return ret;
197 
198 	cs = intel_ring_begin(req, count * 2 + 2);
199 	if (IS_ERR(cs))
200 		return PTR_ERR(cs);
201 
202 	*cs++ = MI_LOAD_REGISTER_IMM(count);
203 	for (mmio = gvt->engine_mmio_list.mmio;
204 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
205 		if (mmio->ring_id != ring_id ||
206 		    !mmio->in_context)
207 			continue;
208 
209 		*cs++ = i915_mmio_reg_offset(mmio->reg);
210 		*cs++ = vgpu_vreg_t(vgpu, mmio->reg) |
211 				(mmio->mask << 16);
212 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
213 			      *(cs-2), *(cs-1), vgpu->id, ring_id);
214 	}
215 
216 	*cs++ = MI_NOOP;
217 	intel_ring_advance(req, cs);
218 
219 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
220 	if (ret)
221 		return ret;
222 
223 	return 0;
224 }
225 
226 static int
227 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
228 					struct i915_request *req)
229 {
230 	unsigned int index;
231 	u32 *cs;
232 
233 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
234 	if (IS_ERR(cs))
235 		return PTR_ERR(cs);
236 
237 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
238 
239 	for (index = 0; index < GEN9_MOCS_SIZE; index++) {
240 		*cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
241 		*cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
242 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
243 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
244 
245 	}
246 
247 	*cs++ = MI_NOOP;
248 	intel_ring_advance(req, cs);
249 
250 	return 0;
251 }
252 
253 static int
254 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
255 				     struct i915_request *req)
256 {
257 	unsigned int index;
258 	u32 *cs;
259 
260 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
261 	if (IS_ERR(cs))
262 		return PTR_ERR(cs);
263 
264 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
265 
266 	for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
267 		*cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
268 		*cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
269 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
270 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
271 
272 	}
273 
274 	*cs++ = MI_NOOP;
275 	intel_ring_advance(req, cs);
276 
277 	return 0;
278 }
279 
280 /*
281  * Use lri command to initialize the mmio which is in context state image for
282  * inhibit context, it contains tracked engine mmio, render_mocs and
283  * render_mocs_l3cc.
284  */
285 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
286 				       struct i915_request *req)
287 {
288 	int ret;
289 	u32 *cs;
290 
291 	cs = intel_ring_begin(req, 2);
292 	if (IS_ERR(cs))
293 		return PTR_ERR(cs);
294 
295 	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
296 	*cs++ = MI_NOOP;
297 	intel_ring_advance(req, cs);
298 
299 	ret = restore_context_mmio_for_inhibit(vgpu, req);
300 	if (ret)
301 		goto out;
302 
303 	/* no MOCS register in context except render engine */
304 	if (req->engine->id != RCS)
305 		goto out;
306 
307 	ret = restore_render_mocs_control_for_inhibit(vgpu, req);
308 	if (ret)
309 		goto out;
310 
311 	ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
312 	if (ret)
313 		goto out;
314 
315 out:
316 	cs = intel_ring_begin(req, 2);
317 	if (IS_ERR(cs))
318 		return PTR_ERR(cs);
319 
320 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
321 	*cs++ = MI_NOOP;
322 	intel_ring_advance(req, cs);
323 
324 	return ret;
325 }
326 
327 static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
328 {
329 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
330 	struct intel_vgpu_submission *s = &vgpu->submission;
331 	enum forcewake_domains fw;
332 	i915_reg_t reg;
333 	u32 regs[] = {
334 		[RCS] = 0x4260,
335 		[VCS] = 0x4264,
336 		[VCS2] = 0x4268,
337 		[BCS] = 0x426c,
338 		[VECS] = 0x4270,
339 	};
340 
341 	if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
342 		return;
343 
344 	if (!test_and_clear_bit(ring_id, (void *)s->tlb_handle_pending))
345 		return;
346 
347 	reg = _MMIO(regs[ring_id]);
348 
349 	/* WaForceWakeRenderDuringMmioTLBInvalidate:skl
350 	 * we need to put a forcewake when invalidating RCS TLB caches,
351 	 * otherwise device can go to RC6 state and interrupt invalidation
352 	 * process
353 	 */
354 	fw = intel_uncore_forcewake_for_reg(dev_priv, reg,
355 					    FW_REG_READ | FW_REG_WRITE);
356 	if (ring_id == RCS && (IS_SKYLAKE(dev_priv) ||
357 			IS_KABYLAKE(dev_priv) || IS_BROXTON(dev_priv)))
358 		fw |= FORCEWAKE_RENDER;
359 
360 	intel_uncore_forcewake_get(dev_priv, fw);
361 
362 	I915_WRITE_FW(reg, 0x1);
363 
364 	if (wait_for_atomic((I915_READ_FW(reg) == 0), 50))
365 		gvt_vgpu_err("timeout in invalidate ring (%d) tlb\n", ring_id);
366 	else
367 		vgpu_vreg_t(vgpu, reg) = 0;
368 
369 	intel_uncore_forcewake_put(dev_priv, fw);
370 
371 	gvt_dbg_core("invalidate TLB for ring %d\n", ring_id);
372 }
373 
374 static void switch_mocs(struct intel_vgpu *pre, struct intel_vgpu *next,
375 			int ring_id)
376 {
377 	struct drm_i915_private *dev_priv;
378 	i915_reg_t offset, l3_offset;
379 	u32 old_v, new_v;
380 
381 	u32 regs[] = {
382 		[RCS] = 0xc800,
383 		[VCS] = 0xc900,
384 		[VCS2] = 0xca00,
385 		[BCS] = 0xcc00,
386 		[VECS] = 0xcb00,
387 	};
388 	int i;
389 
390 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
391 	if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
392 		return;
393 
394 	if ((IS_KABYLAKE(dev_priv) || IS_BROXTON(dev_priv)) && ring_id == RCS)
395 		return;
396 
397 	if (!pre && !gen9_render_mocs.initialized)
398 		load_render_mocs(dev_priv);
399 
400 	offset.reg = regs[ring_id];
401 	for (i = 0; i < GEN9_MOCS_SIZE; i++) {
402 		if (pre)
403 			old_v = vgpu_vreg_t(pre, offset);
404 		else
405 			old_v = gen9_render_mocs.control_table[ring_id][i];
406 		if (next)
407 			new_v = vgpu_vreg_t(next, offset);
408 		else
409 			new_v = gen9_render_mocs.control_table[ring_id][i];
410 
411 		if (old_v != new_v)
412 			I915_WRITE_FW(offset, new_v);
413 
414 		offset.reg += 4;
415 	}
416 
417 	if (ring_id == RCS) {
418 		l3_offset.reg = 0xb020;
419 		for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
420 			if (pre)
421 				old_v = vgpu_vreg_t(pre, l3_offset);
422 			else
423 				old_v = gen9_render_mocs.l3cc_table[i];
424 			if (next)
425 				new_v = vgpu_vreg_t(next, l3_offset);
426 			else
427 				new_v = gen9_render_mocs.l3cc_table[i];
428 
429 			if (old_v != new_v)
430 				I915_WRITE_FW(l3_offset, new_v);
431 
432 			l3_offset.reg += 4;
433 		}
434 	}
435 }
436 
437 #define CTX_CONTEXT_CONTROL_VAL	0x03
438 
439 bool is_inhibit_context(struct intel_context *ce)
440 {
441 	const u32 *reg_state = ce->lrc_reg_state;
442 	u32 inhibit_mask =
443 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
444 
445 	return inhibit_mask ==
446 		(reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
447 }
448 
449 /* Switch ring mmio values (context). */
450 static void switch_mmio(struct intel_vgpu *pre,
451 			struct intel_vgpu *next,
452 			int ring_id)
453 {
454 	struct drm_i915_private *dev_priv;
455 	struct intel_vgpu_submission *s;
456 	struct engine_mmio *mmio;
457 	u32 old_v, new_v;
458 
459 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
460 	if (IS_SKYLAKE(dev_priv)
461 		|| IS_KABYLAKE(dev_priv)
462 		|| IS_BROXTON(dev_priv))
463 		switch_mocs(pre, next, ring_id);
464 
465 	for (mmio = dev_priv->gvt->engine_mmio_list.mmio;
466 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
467 		if (mmio->ring_id != ring_id)
468 			continue;
469 		/*
470 		 * No need to do save or restore of the mmio which is in context
471 		 * state image on kabylake, it's initialized by lri command and
472 		 * save or restore with context together.
473 		 */
474 		if ((IS_KABYLAKE(dev_priv) || IS_BROXTON(dev_priv))
475 			&& mmio->in_context)
476 			continue;
477 
478 		// save
479 		if (pre) {
480 			vgpu_vreg_t(pre, mmio->reg) = I915_READ_FW(mmio->reg);
481 			if (mmio->mask)
482 				vgpu_vreg_t(pre, mmio->reg) &=
483 						~(mmio->mask << 16);
484 			old_v = vgpu_vreg_t(pre, mmio->reg);
485 		} else
486 			old_v = mmio->value = I915_READ_FW(mmio->reg);
487 
488 		// restore
489 		if (next) {
490 			s = &next->submission;
491 			/*
492 			 * No need to restore the mmio which is in context state
493 			 * image if it's not inhibit context, it will restore
494 			 * itself.
495 			 */
496 			if (mmio->in_context &&
497 			    !is_inhibit_context(&s->shadow_ctx->__engine[ring_id]))
498 				continue;
499 
500 			if (mmio->mask)
501 				new_v = vgpu_vreg_t(next, mmio->reg) |
502 							(mmio->mask << 16);
503 			else
504 				new_v = vgpu_vreg_t(next, mmio->reg);
505 		} else {
506 			if (mmio->in_context)
507 				continue;
508 			if (mmio->mask)
509 				new_v = mmio->value | (mmio->mask << 16);
510 			else
511 				new_v = mmio->value;
512 		}
513 
514 		I915_WRITE_FW(mmio->reg, new_v);
515 
516 		trace_render_mmio(pre ? pre->id : 0,
517 				  next ? next->id : 0,
518 				  "switch",
519 				  i915_mmio_reg_offset(mmio->reg),
520 				  old_v, new_v);
521 	}
522 
523 	if (next)
524 		handle_tlb_pending_event(next, ring_id);
525 }
526 
527 /**
528  * intel_gvt_switch_render_mmio - switch mmio context of specific engine
529  * @pre: the last vGPU that own the engine
530  * @next: the vGPU to switch to
531  * @ring_id: specify the engine
532  *
533  * If pre is null indicates that host own the engine. If next is null
534  * indicates that we are switching to host workload.
535  */
536 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
537 			   struct intel_vgpu *next, int ring_id)
538 {
539 	struct drm_i915_private *dev_priv;
540 
541 	if (WARN_ON(!pre && !next))
542 		return;
543 
544 	gvt_dbg_render("switch ring %d from %s to %s\n", ring_id,
545 		       pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
546 
547 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
548 
549 	/**
550 	 * We are using raw mmio access wrapper to improve the
551 	 * performace for batch mmio read/write, so we need
552 	 * handle forcewake mannually.
553 	 */
554 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
555 	switch_mmio(pre, next, ring_id);
556 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
557 }
558 
559 /**
560  * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list
561  * @gvt: GVT device
562  *
563  */
564 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
565 {
566 	struct engine_mmio *mmio;
567 
568 	if (IS_SKYLAKE(gvt->dev_priv) ||
569 		IS_KABYLAKE(gvt->dev_priv) ||
570 		IS_BROXTON(gvt->dev_priv))
571 		gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
572 	else
573 		gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
574 
575 	for (mmio = gvt->engine_mmio_list.mmio;
576 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
577 		if (mmio->in_context) {
578 			gvt->engine_mmio_list.ctx_mmio_count[mmio->ring_id]++;
579 			intel_gvt_mmio_set_in_ctx(gvt, mmio->reg.reg);
580 		}
581 	}
582 }
583