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 	{RCS, _MMIO(0x20D8), 0xffff, true}, /* 0x20d8 */
136 
137 	{RCS, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */
138 	{RCS, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */
139 	{RCS, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */
140 	{RCS, INVALID_MMIO_REG, 0, false } /* Terminated */
141 };
142 
143 static struct {
144 	bool initialized;
145 	u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
146 	u32 l3cc_table[GEN9_MOCS_SIZE / 2];
147 } gen9_render_mocs;
148 
149 static void load_render_mocs(struct drm_i915_private *dev_priv)
150 {
151 	i915_reg_t offset;
152 	u32 regs[] = {
153 		[RCS] = 0xc800,
154 		[VCS] = 0xc900,
155 		[VCS2] = 0xca00,
156 		[BCS] = 0xcc00,
157 		[VECS] = 0xcb00,
158 	};
159 	int ring_id, i;
160 
161 	for (ring_id = 0; ring_id < ARRAY_SIZE(regs); ring_id++) {
162 		if (!HAS_ENGINE(dev_priv, ring_id))
163 			continue;
164 		offset.reg = regs[ring_id];
165 		for (i = 0; i < GEN9_MOCS_SIZE; i++) {
166 			gen9_render_mocs.control_table[ring_id][i] =
167 				I915_READ_FW(offset);
168 			offset.reg += 4;
169 		}
170 	}
171 
172 	offset.reg = 0xb020;
173 	for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
174 		gen9_render_mocs.l3cc_table[i] =
175 			I915_READ_FW(offset);
176 		offset.reg += 4;
177 	}
178 	gen9_render_mocs.initialized = true;
179 }
180 
181 static int
182 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
183 				 struct i915_request *req)
184 {
185 	u32 *cs;
186 	int ret;
187 	struct engine_mmio *mmio;
188 	struct intel_gvt *gvt = vgpu->gvt;
189 	int ring_id = req->engine->id;
190 	int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
191 
192 	if (count == 0)
193 		return 0;
194 
195 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
196 	if (ret)
197 		return ret;
198 
199 	cs = intel_ring_begin(req, count * 2 + 2);
200 	if (IS_ERR(cs))
201 		return PTR_ERR(cs);
202 
203 	*cs++ = MI_LOAD_REGISTER_IMM(count);
204 	for (mmio = gvt->engine_mmio_list.mmio;
205 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
206 		if (mmio->ring_id != ring_id ||
207 		    !mmio->in_context)
208 			continue;
209 
210 		*cs++ = i915_mmio_reg_offset(mmio->reg);
211 		*cs++ = vgpu_vreg_t(vgpu, mmio->reg) |
212 				(mmio->mask << 16);
213 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
214 			      *(cs-2), *(cs-1), vgpu->id, ring_id);
215 	}
216 
217 	*cs++ = MI_NOOP;
218 	intel_ring_advance(req, cs);
219 
220 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
221 	if (ret)
222 		return ret;
223 
224 	return 0;
225 }
226 
227 static int
228 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
229 					struct i915_request *req)
230 {
231 	unsigned int index;
232 	u32 *cs;
233 
234 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
235 	if (IS_ERR(cs))
236 		return PTR_ERR(cs);
237 
238 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
239 
240 	for (index = 0; index < GEN9_MOCS_SIZE; index++) {
241 		*cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
242 		*cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
243 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
244 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
245 
246 	}
247 
248 	*cs++ = MI_NOOP;
249 	intel_ring_advance(req, cs);
250 
251 	return 0;
252 }
253 
254 static int
255 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
256 				     struct i915_request *req)
257 {
258 	unsigned int index;
259 	u32 *cs;
260 
261 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
262 	if (IS_ERR(cs))
263 		return PTR_ERR(cs);
264 
265 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
266 
267 	for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
268 		*cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
269 		*cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
270 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
271 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
272 
273 	}
274 
275 	*cs++ = MI_NOOP;
276 	intel_ring_advance(req, cs);
277 
278 	return 0;
279 }
280 
281 /*
282  * Use lri command to initialize the mmio which is in context state image for
283  * inhibit context, it contains tracked engine mmio, render_mocs and
284  * render_mocs_l3cc.
285  */
286 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
287 				       struct i915_request *req)
288 {
289 	int ret;
290 	u32 *cs;
291 
292 	cs = intel_ring_begin(req, 2);
293 	if (IS_ERR(cs))
294 		return PTR_ERR(cs);
295 
296 	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
297 	*cs++ = MI_NOOP;
298 	intel_ring_advance(req, cs);
299 
300 	ret = restore_context_mmio_for_inhibit(vgpu, req);
301 	if (ret)
302 		goto out;
303 
304 	/* no MOCS register in context except render engine */
305 	if (req->engine->id != RCS)
306 		goto out;
307 
308 	ret = restore_render_mocs_control_for_inhibit(vgpu, req);
309 	if (ret)
310 		goto out;
311 
312 	ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
313 	if (ret)
314 		goto out;
315 
316 out:
317 	cs = intel_ring_begin(req, 2);
318 	if (IS_ERR(cs))
319 		return PTR_ERR(cs);
320 
321 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
322 	*cs++ = MI_NOOP;
323 	intel_ring_advance(req, cs);
324 
325 	return ret;
326 }
327 
328 static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
329 {
330 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
331 	struct intel_vgpu_submission *s = &vgpu->submission;
332 	enum forcewake_domains fw;
333 	i915_reg_t reg;
334 	u32 regs[] = {
335 		[RCS] = 0x4260,
336 		[VCS] = 0x4264,
337 		[VCS2] = 0x4268,
338 		[BCS] = 0x426c,
339 		[VECS] = 0x4270,
340 	};
341 
342 	if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
343 		return;
344 
345 	if (!test_and_clear_bit(ring_id, (void *)s->tlb_handle_pending))
346 		return;
347 
348 	reg = _MMIO(regs[ring_id]);
349 
350 	/* WaForceWakeRenderDuringMmioTLBInvalidate:skl
351 	 * we need to put a forcewake when invalidating RCS TLB caches,
352 	 * otherwise device can go to RC6 state and interrupt invalidation
353 	 * process
354 	 */
355 	fw = intel_uncore_forcewake_for_reg(dev_priv, reg,
356 					    FW_REG_READ | FW_REG_WRITE);
357 	if (ring_id == RCS && (INTEL_GEN(dev_priv) >= 9))
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)
395 		|| IS_COFFEELAKE(dev_priv)) && ring_id == RCS)
396 		return;
397 
398 	if (!pre && !gen9_render_mocs.initialized)
399 		load_render_mocs(dev_priv);
400 
401 	offset.reg = regs[ring_id];
402 	for (i = 0; i < GEN9_MOCS_SIZE; i++) {
403 		if (pre)
404 			old_v = vgpu_vreg_t(pre, offset);
405 		else
406 			old_v = gen9_render_mocs.control_table[ring_id][i];
407 		if (next)
408 			new_v = vgpu_vreg_t(next, offset);
409 		else
410 			new_v = gen9_render_mocs.control_table[ring_id][i];
411 
412 		if (old_v != new_v)
413 			I915_WRITE_FW(offset, new_v);
414 
415 		offset.reg += 4;
416 	}
417 
418 	if (ring_id == RCS) {
419 		l3_offset.reg = 0xb020;
420 		for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
421 			if (pre)
422 				old_v = vgpu_vreg_t(pre, l3_offset);
423 			else
424 				old_v = gen9_render_mocs.l3cc_table[i];
425 			if (next)
426 				new_v = vgpu_vreg_t(next, l3_offset);
427 			else
428 				new_v = gen9_render_mocs.l3cc_table[i];
429 
430 			if (old_v != new_v)
431 				I915_WRITE_FW(l3_offset, new_v);
432 
433 			l3_offset.reg += 4;
434 		}
435 	}
436 }
437 
438 #define CTX_CONTEXT_CONTROL_VAL	0x03
439 
440 bool is_inhibit_context(struct intel_context *ce)
441 {
442 	const u32 *reg_state = ce->lrc_reg_state;
443 	u32 inhibit_mask =
444 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
445 
446 	return inhibit_mask ==
447 		(reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
448 }
449 
450 /* Switch ring mmio values (context). */
451 static void switch_mmio(struct intel_vgpu *pre,
452 			struct intel_vgpu *next,
453 			int ring_id)
454 {
455 	struct drm_i915_private *dev_priv;
456 	struct intel_vgpu_submission *s;
457 	struct engine_mmio *mmio;
458 	u32 old_v, new_v;
459 
460 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
461 	if (INTEL_GEN(dev_priv) >= 9)
462 		switch_mocs(pre, next, ring_id);
463 
464 	for (mmio = dev_priv->gvt->engine_mmio_list.mmio;
465 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
466 		if (mmio->ring_id != ring_id)
467 			continue;
468 		/*
469 		 * No need to do save or restore of the mmio which is in context
470 		 * state image on kabylake, it's initialized by lri command and
471 		 * save or restore with context together.
472 		 */
473 		if ((IS_KABYLAKE(dev_priv) || IS_BROXTON(dev_priv)
474 			|| IS_COFFEELAKE(dev_priv)) && mmio->in_context)
475 			continue;
476 
477 		// save
478 		if (pre) {
479 			vgpu_vreg_t(pre, mmio->reg) = I915_READ_FW(mmio->reg);
480 			if (mmio->mask)
481 				vgpu_vreg_t(pre, mmio->reg) &=
482 						~(mmio->mask << 16);
483 			old_v = vgpu_vreg_t(pre, mmio->reg);
484 		} else
485 			old_v = mmio->value = I915_READ_FW(mmio->reg);
486 
487 		// restore
488 		if (next) {
489 			s = &next->submission;
490 			/*
491 			 * No need to restore the mmio which is in context state
492 			 * image if it's not inhibit context, it will restore
493 			 * itself.
494 			 */
495 			if (mmio->in_context &&
496 			    !is_inhibit_context(&s->shadow_ctx->__engine[ring_id]))
497 				continue;
498 
499 			if (mmio->mask)
500 				new_v = vgpu_vreg_t(next, mmio->reg) |
501 							(mmio->mask << 16);
502 			else
503 				new_v = vgpu_vreg_t(next, mmio->reg);
504 		} else {
505 			if (mmio->in_context)
506 				continue;
507 			if (mmio->mask)
508 				new_v = mmio->value | (mmio->mask << 16);
509 			else
510 				new_v = mmio->value;
511 		}
512 
513 		I915_WRITE_FW(mmio->reg, new_v);
514 
515 		trace_render_mmio(pre ? pre->id : 0,
516 				  next ? next->id : 0,
517 				  "switch",
518 				  i915_mmio_reg_offset(mmio->reg),
519 				  old_v, new_v);
520 	}
521 
522 	if (next)
523 		handle_tlb_pending_event(next, ring_id);
524 }
525 
526 /**
527  * intel_gvt_switch_render_mmio - switch mmio context of specific engine
528  * @pre: the last vGPU that own the engine
529  * @next: the vGPU to switch to
530  * @ring_id: specify the engine
531  *
532  * If pre is null indicates that host own the engine. If next is null
533  * indicates that we are switching to host workload.
534  */
535 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
536 			   struct intel_vgpu *next, int ring_id)
537 {
538 	struct drm_i915_private *dev_priv;
539 
540 	if (WARN_ON(!pre && !next))
541 		return;
542 
543 	gvt_dbg_render("switch ring %d from %s to %s\n", ring_id,
544 		       pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
545 
546 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
547 
548 	/**
549 	 * We are using raw mmio access wrapper to improve the
550 	 * performace for batch mmio read/write, so we need
551 	 * handle forcewake mannually.
552 	 */
553 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
554 	switch_mmio(pre, next, ring_id);
555 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
556 }
557 
558 /**
559  * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list
560  * @gvt: GVT device
561  *
562  */
563 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
564 {
565 	struct engine_mmio *mmio;
566 
567 	if (INTEL_GEN(gvt->dev_priv) >= 9)
568 		gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
569 	else
570 		gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
571 
572 	for (mmio = gvt->engine_mmio_list.mmio;
573 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
574 		if (mmio->in_context) {
575 			gvt->engine_mmio_list.ctx_mmio_count[mmio->ring_id]++;
576 			intel_gvt_mmio_set_in_ctx(gvt, mmio->reg.reg);
577 		}
578 	}
579 }
580