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 /**
41  * Defined in Intel Open Source PRM.
42  * Ref: https://01.org/linuxgraphics/documentation/hardware-specification-prms
43  */
44 #define TRVATTL3PTRDW(i)	_MMIO(0x4de0 + (i)*4)
45 #define TRNULLDETCT		_MMIO(0x4de8)
46 #define TRINVTILEDETCT		_MMIO(0x4dec)
47 #define TRVADR			_MMIO(0x4df0)
48 #define TRTTE			_MMIO(0x4df4)
49 #define RING_EXCC(base)		_MMIO((base) + 0x28)
50 #define RING_GFX_MODE(base)	_MMIO((base) + 0x29c)
51 #define VF_GUARDBAND		_MMIO(0x83a4)
52 
53 #define GEN9_MOCS_SIZE		64
54 
55 /* Raw offset is appened to each line for convenience. */
56 static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = {
57 	{RCS, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
58 	{RCS, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
59 	{RCS, HWSTAM, 0x0, false}, /* 0x2098 */
60 	{RCS, INSTPM, 0xffff, true}, /* 0x20c0 */
61 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
62 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
63 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
64 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
65 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
66 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
67 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
68 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
69 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
70 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
71 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
72 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
73 	{RCS, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
74 	{RCS, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
75 	{RCS, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
76 	{RCS, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
77 	{RCS, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
78 	{RCS, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
79 
80 	{BCS, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
81 	{BCS, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
82 	{BCS, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
83 	{BCS, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
84 	{BCS, RING_EXCC(BLT_RING_BASE), 0x0, false}, /* 0x22028 */
85 	{RCS, INVALID_MMIO_REG, 0, false } /* Terminated */
86 };
87 
88 static struct engine_mmio gen9_engine_mmio_list[] __cacheline_aligned = {
89 	{RCS, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
90 	{RCS, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
91 	{RCS, HWSTAM, 0x0, false}, /* 0x2098 */
92 	{RCS, INSTPM, 0xffff, true}, /* 0x20c0 */
93 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
94 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
95 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
96 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
97 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
98 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
99 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
100 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
101 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
102 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
103 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
104 	{RCS, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
105 	{RCS, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
106 	{RCS, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
107 	{RCS, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
108 	{RCS, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
109 	{RCS, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
110 	{RCS, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
111 
112 	{RCS, GEN8_PRIVATE_PAT_LO, 0, false}, /* 0x40e0 */
113 	{RCS, GEN8_PRIVATE_PAT_HI, 0, false}, /* 0x40e4 */
114 	{RCS, GEN8_CS_CHICKEN1, 0xffff, true}, /* 0x2580 */
115 	{RCS, COMMON_SLICE_CHICKEN2, 0xffff, true}, /* 0x7014 */
116 	{RCS, GEN9_CS_DEBUG_MODE1, 0xffff, false}, /* 0x20ec */
117 	{RCS, GEN8_L3SQCREG4, 0, false}, /* 0xb118 */
118 	{RCS, GEN7_HALF_SLICE_CHICKEN1, 0xffff, true}, /* 0xe100 */
119 	{RCS, HALF_SLICE_CHICKEN2, 0xffff, true}, /* 0xe180 */
120 	{RCS, HALF_SLICE_CHICKEN3, 0xffff, true}, /* 0xe184 */
121 	{RCS, GEN9_HALF_SLICE_CHICKEN5, 0xffff, true}, /* 0xe188 */
122 	{RCS, GEN9_HALF_SLICE_CHICKEN7, 0xffff, true}, /* 0xe194 */
123 	{RCS, TRVATTL3PTRDW(0), 0, false}, /* 0x4de0 */
124 	{RCS, TRVATTL3PTRDW(1), 0, false}, /* 0x4de4 */
125 	{RCS, TRNULLDETCT, 0, false}, /* 0x4de8 */
126 	{RCS, TRINVTILEDETCT, 0, false}, /* 0x4dec */
127 	{RCS, TRVADR, 0, false}, /* 0x4df0 */
128 	{RCS, TRTTE, 0, false}, /* 0x4df4 */
129 
130 	{BCS, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
131 	{BCS, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
132 	{BCS, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
133 	{BCS, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
134 	{BCS, RING_EXCC(BLT_RING_BASE), 0x0, false}, /* 0x22028 */
135 
136 	{VCS2, RING_EXCC(GEN8_BSD2_RING_BASE), 0xffff, false}, /* 0x1c028 */
137 
138 	{VECS, RING_EXCC(VEBOX_RING_BASE), 0xffff, false}, /* 0x1a028 */
139 
140 	{RCS, GEN8_HDC_CHICKEN1, 0xffff, true}, /* 0x7304 */
141 	{RCS, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
142 	{RCS, GEN7_UCGCTL4, 0x0, false}, /* 0x940c */
143 	{RCS, GAMT_CHKN_BIT_REG, 0x0, false}, /* 0x4ab8 */
144 
145 	{RCS, GEN9_GAMT_ECO_REG_RW_IA, 0x0, false}, /* 0x4ab0 */
146 	{RCS, GEN9_CSFE_CHICKEN1_RCS, 0x0, false}, /* 0x20d4 */
147 
148 	{RCS, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */
149 	{RCS, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */
150 	{RCS, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */
151 	{RCS, INVALID_MMIO_REG, 0, false } /* Terminated */
152 };
153 
154 static struct {
155 	bool initialized;
156 	u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
157 	u32 l3cc_table[GEN9_MOCS_SIZE / 2];
158 } gen9_render_mocs;
159 
160 static void load_render_mocs(struct drm_i915_private *dev_priv)
161 {
162 	i915_reg_t offset;
163 	u32 regs[] = {
164 		[RCS] = 0xc800,
165 		[VCS] = 0xc900,
166 		[VCS2] = 0xca00,
167 		[BCS] = 0xcc00,
168 		[VECS] = 0xcb00,
169 	};
170 	int ring_id, i;
171 
172 	for (ring_id = 0; ring_id < ARRAY_SIZE(regs); ring_id++) {
173 		offset.reg = regs[ring_id];
174 		for (i = 0; i < GEN9_MOCS_SIZE; i++) {
175 			gen9_render_mocs.control_table[ring_id][i] =
176 				I915_READ_FW(offset);
177 			offset.reg += 4;
178 		}
179 	}
180 
181 	offset.reg = 0xb020;
182 	for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
183 		gen9_render_mocs.l3cc_table[i] =
184 			I915_READ_FW(offset);
185 		offset.reg += 4;
186 	}
187 	gen9_render_mocs.initialized = true;
188 }
189 
190 static int
191 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
192 				 struct i915_request *req)
193 {
194 	u32 *cs;
195 	int ret;
196 	struct engine_mmio *mmio;
197 	struct intel_gvt *gvt = vgpu->gvt;
198 	int ring_id = req->engine->id;
199 	int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
200 
201 	if (count == 0)
202 		return 0;
203 
204 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
205 	if (ret)
206 		return ret;
207 
208 	cs = intel_ring_begin(req, count * 2 + 2);
209 	if (IS_ERR(cs))
210 		return PTR_ERR(cs);
211 
212 	*cs++ = MI_LOAD_REGISTER_IMM(count);
213 	for (mmio = gvt->engine_mmio_list.mmio;
214 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
215 		if (mmio->ring_id != ring_id ||
216 		    !mmio->in_context)
217 			continue;
218 
219 		*cs++ = i915_mmio_reg_offset(mmio->reg);
220 		*cs++ = vgpu_vreg_t(vgpu, mmio->reg) |
221 				(mmio->mask << 16);
222 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
223 			      *(cs-2), *(cs-1), vgpu->id, ring_id);
224 	}
225 
226 	*cs++ = MI_NOOP;
227 	intel_ring_advance(req, cs);
228 
229 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
230 	if (ret)
231 		return ret;
232 
233 	return 0;
234 }
235 
236 static int
237 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
238 					struct i915_request *req)
239 {
240 	unsigned int index;
241 	u32 *cs;
242 
243 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
244 	if (IS_ERR(cs))
245 		return PTR_ERR(cs);
246 
247 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
248 
249 	for (index = 0; index < GEN9_MOCS_SIZE; index++) {
250 		*cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
251 		*cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
252 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
253 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
254 
255 	}
256 
257 	*cs++ = MI_NOOP;
258 	intel_ring_advance(req, cs);
259 
260 	return 0;
261 }
262 
263 static int
264 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
265 				     struct i915_request *req)
266 {
267 	unsigned int index;
268 	u32 *cs;
269 
270 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
271 	if (IS_ERR(cs))
272 		return PTR_ERR(cs);
273 
274 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
275 
276 	for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
277 		*cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
278 		*cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
279 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
280 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
281 
282 	}
283 
284 	*cs++ = MI_NOOP;
285 	intel_ring_advance(req, cs);
286 
287 	return 0;
288 }
289 
290 /*
291  * Use lri command to initialize the mmio which is in context state image for
292  * inhibit context, it contains tracked engine mmio, render_mocs and
293  * render_mocs_l3cc.
294  */
295 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
296 				       struct i915_request *req)
297 {
298 	int ret;
299 	u32 *cs;
300 
301 	cs = intel_ring_begin(req, 2);
302 	if (IS_ERR(cs))
303 		return PTR_ERR(cs);
304 
305 	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
306 	*cs++ = MI_NOOP;
307 	intel_ring_advance(req, cs);
308 
309 	ret = restore_context_mmio_for_inhibit(vgpu, req);
310 	if (ret)
311 		goto out;
312 
313 	/* no MOCS register in context except render engine */
314 	if (req->engine->id != RCS)
315 		goto out;
316 
317 	ret = restore_render_mocs_control_for_inhibit(vgpu, req);
318 	if (ret)
319 		goto out;
320 
321 	ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
322 	if (ret)
323 		goto out;
324 
325 out:
326 	cs = intel_ring_begin(req, 2);
327 	if (IS_ERR(cs))
328 		return PTR_ERR(cs);
329 
330 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
331 	*cs++ = MI_NOOP;
332 	intel_ring_advance(req, cs);
333 
334 	return ret;
335 }
336 
337 static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
338 {
339 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
340 	struct intel_vgpu_submission *s = &vgpu->submission;
341 	enum forcewake_domains fw;
342 	i915_reg_t reg;
343 	u32 regs[] = {
344 		[RCS] = 0x4260,
345 		[VCS] = 0x4264,
346 		[VCS2] = 0x4268,
347 		[BCS] = 0x426c,
348 		[VECS] = 0x4270,
349 	};
350 
351 	if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
352 		return;
353 
354 	if (!test_and_clear_bit(ring_id, (void *)s->tlb_handle_pending))
355 		return;
356 
357 	reg = _MMIO(regs[ring_id]);
358 
359 	/* WaForceWakeRenderDuringMmioTLBInvalidate:skl
360 	 * we need to put a forcewake when invalidating RCS TLB caches,
361 	 * otherwise device can go to RC6 state and interrupt invalidation
362 	 * process
363 	 */
364 	fw = intel_uncore_forcewake_for_reg(dev_priv, reg,
365 					    FW_REG_READ | FW_REG_WRITE);
366 	if (ring_id == RCS && (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)))
367 		fw |= FORCEWAKE_RENDER;
368 
369 	intel_uncore_forcewake_get(dev_priv, fw);
370 
371 	I915_WRITE_FW(reg, 0x1);
372 
373 	if (wait_for_atomic((I915_READ_FW(reg) == 0), 50))
374 		gvt_vgpu_err("timeout in invalidate ring (%d) tlb\n", ring_id);
375 	else
376 		vgpu_vreg_t(vgpu, reg) = 0;
377 
378 	intel_uncore_forcewake_put(dev_priv, fw);
379 
380 	gvt_dbg_core("invalidate TLB for ring %d\n", ring_id);
381 }
382 
383 static void switch_mocs(struct intel_vgpu *pre, struct intel_vgpu *next,
384 			int ring_id)
385 {
386 	struct drm_i915_private *dev_priv;
387 	i915_reg_t offset, l3_offset;
388 	u32 old_v, new_v;
389 
390 	u32 regs[] = {
391 		[RCS] = 0xc800,
392 		[VCS] = 0xc900,
393 		[VCS2] = 0xca00,
394 		[BCS] = 0xcc00,
395 		[VECS] = 0xcb00,
396 	};
397 	int i;
398 
399 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
400 	if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
401 		return;
402 
403 	if (IS_KABYLAKE(dev_priv) && ring_id == RCS)
404 		return;
405 
406 	if (!pre && !gen9_render_mocs.initialized)
407 		load_render_mocs(dev_priv);
408 
409 	offset.reg = regs[ring_id];
410 	for (i = 0; i < GEN9_MOCS_SIZE; i++) {
411 		if (pre)
412 			old_v = vgpu_vreg_t(pre, offset);
413 		else
414 			old_v = gen9_render_mocs.control_table[ring_id][i];
415 		if (next)
416 			new_v = vgpu_vreg_t(next, offset);
417 		else
418 			new_v = gen9_render_mocs.control_table[ring_id][i];
419 
420 		if (old_v != new_v)
421 			I915_WRITE_FW(offset, new_v);
422 
423 		offset.reg += 4;
424 	}
425 
426 	if (ring_id == RCS) {
427 		l3_offset.reg = 0xb020;
428 		for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
429 			if (pre)
430 				old_v = vgpu_vreg_t(pre, l3_offset);
431 			else
432 				old_v = gen9_render_mocs.l3cc_table[i];
433 			if (next)
434 				new_v = vgpu_vreg_t(next, l3_offset);
435 			else
436 				new_v = gen9_render_mocs.l3cc_table[i];
437 
438 			if (old_v != new_v)
439 				I915_WRITE_FW(l3_offset, new_v);
440 
441 			l3_offset.reg += 4;
442 		}
443 	}
444 }
445 
446 #define CTX_CONTEXT_CONTROL_VAL	0x03
447 
448 bool is_inhibit_context(struct i915_gem_context *ctx, int ring_id)
449 {
450 	u32 *reg_state = ctx->engine[ring_id].lrc_reg_state;
451 	u32 inhibit_mask =
452 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
453 
454 	return inhibit_mask ==
455 		(reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
456 }
457 
458 /* Switch ring mmio values (context). */
459 static void switch_mmio(struct intel_vgpu *pre,
460 			struct intel_vgpu *next,
461 			int ring_id)
462 {
463 	struct drm_i915_private *dev_priv;
464 	struct intel_vgpu_submission *s;
465 	struct engine_mmio *mmio;
466 	u32 old_v, new_v;
467 
468 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
469 	if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
470 		switch_mocs(pre, next, ring_id);
471 
472 	for (mmio = dev_priv->gvt->engine_mmio_list.mmio;
473 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
474 		if (mmio->ring_id != ring_id)
475 			continue;
476 		/*
477 		 * No need to do save or restore of the mmio which is in context
478 		 * state image on kabylake, it's initialized by lri command and
479 		 * save or restore with context together.
480 		 */
481 		if (IS_KABYLAKE(dev_priv) && mmio->in_context)
482 			continue;
483 
484 		// save
485 		if (pre) {
486 			vgpu_vreg_t(pre, mmio->reg) = I915_READ_FW(mmio->reg);
487 			if (mmio->mask)
488 				vgpu_vreg_t(pre, mmio->reg) &=
489 						~(mmio->mask << 16);
490 			old_v = vgpu_vreg_t(pre, mmio->reg);
491 		} else
492 			old_v = mmio->value = I915_READ_FW(mmio->reg);
493 
494 		// restore
495 		if (next) {
496 			s = &next->submission;
497 			/*
498 			 * No need to restore the mmio which is in context state
499 			 * image if it's not inhibit context, it will restore
500 			 * itself.
501 			 */
502 			if (mmio->in_context &&
503 			    !is_inhibit_context(s->shadow_ctx, ring_id))
504 				continue;
505 
506 			if (mmio->mask)
507 				new_v = vgpu_vreg_t(next, mmio->reg) |
508 							(mmio->mask << 16);
509 			else
510 				new_v = vgpu_vreg_t(next, mmio->reg);
511 		} else {
512 			if (mmio->in_context)
513 				continue;
514 			if (mmio->mask)
515 				new_v = mmio->value | (mmio->mask << 16);
516 			else
517 				new_v = mmio->value;
518 		}
519 
520 		I915_WRITE_FW(mmio->reg, new_v);
521 
522 		trace_render_mmio(pre ? pre->id : 0,
523 				  next ? next->id : 0,
524 				  "switch",
525 				  i915_mmio_reg_offset(mmio->reg),
526 				  old_v, new_v);
527 	}
528 
529 	if (next)
530 		handle_tlb_pending_event(next, ring_id);
531 }
532 
533 /**
534  * intel_gvt_switch_render_mmio - switch mmio context of specific engine
535  * @pre: the last vGPU that own the engine
536  * @next: the vGPU to switch to
537  * @ring_id: specify the engine
538  *
539  * If pre is null indicates that host own the engine. If next is null
540  * indicates that we are switching to host workload.
541  */
542 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
543 			   struct intel_vgpu *next, int ring_id)
544 {
545 	struct drm_i915_private *dev_priv;
546 
547 	if (WARN_ON(!pre && !next))
548 		return;
549 
550 	gvt_dbg_render("switch ring %d from %s to %s\n", ring_id,
551 		       pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
552 
553 	dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv;
554 
555 	/**
556 	 * We are using raw mmio access wrapper to improve the
557 	 * performace for batch mmio read/write, so we need
558 	 * handle forcewake mannually.
559 	 */
560 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
561 	switch_mmio(pre, next, ring_id);
562 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
563 }
564 
565 /**
566  * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list
567  * @gvt: GVT device
568  *
569  */
570 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
571 {
572 	struct engine_mmio *mmio;
573 
574 	if (IS_SKYLAKE(gvt->dev_priv) || IS_KABYLAKE(gvt->dev_priv))
575 		gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
576 	else
577 		gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
578 
579 	for (mmio = gvt->engine_mmio_list.mmio;
580 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
581 		if (mmio->in_context)
582 			gvt->engine_mmio_list.ctx_mmio_count[mmio->ring_id]++;
583 	}
584 }
585