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 "gt/intel_context.h"
38 #include "gt/intel_gpu_commands.h"
39 #include "gt/intel_ring.h"
40 #include "gvt.h"
41 #include "trace.h"
42 
43 #define GEN9_MOCS_SIZE		64
44 
45 /* Raw offset is appened to each line for convenience. */
46 static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = {
47 	{RCS0, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
48 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
49 	{RCS0, HWSTAM, 0x0, false}, /* 0x2098 */
50 	{RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */
51 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
52 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
53 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
54 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
55 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
56 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
57 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
58 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
59 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
60 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
61 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
62 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
63 	{RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
64 	{RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
65 	{RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
66 	{RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
67 	{RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
68 	{RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
69 
70 	{BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
71 	{BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
72 	{BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
73 	{BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
74 	{BCS0, RING_EXCC(BLT_RING_BASE), 0xffff, false}, /* 0x22028 */
75 	{RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */
76 };
77 
78 static struct engine_mmio gen9_engine_mmio_list[] __cacheline_aligned = {
79 	{RCS0, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
80 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
81 	{RCS0, HWSTAM, 0x0, false}, /* 0x2098 */
82 	{RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */
83 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
84 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
85 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
86 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
87 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
88 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
89 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
90 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
91 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
92 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
93 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
94 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
95 	{RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
96 	{RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
97 	{RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
98 	{RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
99 	{RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
100 	{RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
101 
102 	{RCS0, GEN8_PRIVATE_PAT_LO, 0, false}, /* 0x40e0 */
103 	{RCS0, GEN8_PRIVATE_PAT_HI, 0, false}, /* 0x40e4 */
104 	{RCS0, GEN8_CS_CHICKEN1, 0xffff, true}, /* 0x2580 */
105 	{RCS0, COMMON_SLICE_CHICKEN2, 0xffff, true}, /* 0x7014 */
106 	{RCS0, GEN9_CS_DEBUG_MODE1, 0xffff, false}, /* 0x20ec */
107 	{RCS0, GEN8_L3SQCREG4, 0, false}, /* 0xb118 */
108 	{RCS0, GEN9_SCRATCH1, 0, false}, /* 0xb11c */
109 	{RCS0, GEN9_SCRATCH_LNCF1, 0, false}, /* 0xb008 */
110 	{RCS0, GEN7_HALF_SLICE_CHICKEN1, 0xffff, true}, /* 0xe100 */
111 	{RCS0, HALF_SLICE_CHICKEN2, 0xffff, true}, /* 0xe180 */
112 	{RCS0, HALF_SLICE_CHICKEN3, 0xffff, true}, /* 0xe184 */
113 	{RCS0, GEN9_HALF_SLICE_CHICKEN5, 0xffff, true}, /* 0xe188 */
114 	{RCS0, GEN9_HALF_SLICE_CHICKEN7, 0xffff, true}, /* 0xe194 */
115 	{RCS0, GEN8_ROW_CHICKEN, 0xffff, true}, /* 0xe4f0 */
116 	{RCS0, TRVATTL3PTRDW(0), 0, true}, /* 0x4de0 */
117 	{RCS0, TRVATTL3PTRDW(1), 0, true}, /* 0x4de4 */
118 	{RCS0, TRNULLDETCT, 0, true}, /* 0x4de8 */
119 	{RCS0, TRINVTILEDETCT, 0, true}, /* 0x4dec */
120 	{RCS0, TRVADR, 0, true}, /* 0x4df0 */
121 	{RCS0, TRTTE, 0, true}, /* 0x4df4 */
122 	{RCS0, _MMIO(0x4dfc), 0, true},
123 
124 	{BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
125 	{BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
126 	{BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
127 	{BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
128 	{BCS0, RING_EXCC(BLT_RING_BASE), 0xffff, false}, /* 0x22028 */
129 
130 	{VCS1, RING_EXCC(GEN8_BSD2_RING_BASE), 0xffff, false}, /* 0x1c028 */
131 
132 	{VECS0, RING_EXCC(VEBOX_RING_BASE), 0xffff, false}, /* 0x1a028 */
133 
134 	{RCS0, GEN8_HDC_CHICKEN1, 0xffff, true}, /* 0x7304 */
135 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
136 	{RCS0, GEN7_UCGCTL4, 0x0, false}, /* 0x940c */
137 	{RCS0, GAMT_CHKN_BIT_REG, 0x0, false}, /* 0x4ab8 */
138 
139 	{RCS0, GEN9_GAMT_ECO_REG_RW_IA, 0x0, false}, /* 0x4ab0 */
140 	{RCS0, GEN9_CSFE_CHICKEN1_RCS, 0xffff, false}, /* 0x20d4 */
141 	{RCS0, _MMIO(0x20D8), 0xffff, true}, /* 0x20d8 */
142 
143 	{RCS0, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */
144 	{RCS0, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */
145 	{RCS0, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */
146 	{RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */
147 };
148 
149 static struct {
150 	bool initialized;
151 	u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
152 	u32 l3cc_table[GEN9_MOCS_SIZE / 2];
153 } gen9_render_mocs;
154 
155 static u32 gen9_mocs_mmio_offset_list[] = {
156 	[RCS0]  = 0xc800,
157 	[VCS0]  = 0xc900,
158 	[VCS1]  = 0xca00,
159 	[BCS0]  = 0xcc00,
160 	[VECS0] = 0xcb00,
161 };
162 
163 static void load_render_mocs(const struct intel_engine_cs *engine)
164 {
165 	struct intel_gvt *gvt = engine->i915->gvt;
166 	struct intel_uncore *uncore = engine->uncore;
167 	u32 cnt = gvt->engine_mmio_list.mocs_mmio_offset_list_cnt;
168 	u32 *regs = gvt->engine_mmio_list.mocs_mmio_offset_list;
169 	i915_reg_t offset;
170 	int ring_id, i;
171 
172 	/* Platform doesn't have mocs mmios. */
173 	if (!regs)
174 		return;
175 
176 	for (ring_id = 0; ring_id < cnt; ring_id++) {
177 		if (!HAS_ENGINE(engine->gt, ring_id))
178 			continue;
179 
180 		offset.reg = regs[ring_id];
181 		for (i = 0; i < GEN9_MOCS_SIZE; i++) {
182 			gen9_render_mocs.control_table[ring_id][i] =
183 				intel_uncore_read_fw(uncore, offset);
184 			offset.reg += 4;
185 		}
186 	}
187 
188 	offset.reg = 0xb020;
189 	for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
190 		gen9_render_mocs.l3cc_table[i] =
191 			intel_uncore_read_fw(uncore, offset);
192 		offset.reg += 4;
193 	}
194 	gen9_render_mocs.initialized = true;
195 }
196 
197 static int
198 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
199 				 struct i915_request *req)
200 {
201 	u32 *cs;
202 	int ret;
203 	struct engine_mmio *mmio;
204 	struct intel_gvt *gvt = vgpu->gvt;
205 	int ring_id = req->engine->id;
206 	int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
207 
208 	if (count == 0)
209 		return 0;
210 
211 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
212 	if (ret)
213 		return ret;
214 
215 	cs = intel_ring_begin(req, count * 2 + 2);
216 	if (IS_ERR(cs))
217 		return PTR_ERR(cs);
218 
219 	*cs++ = MI_LOAD_REGISTER_IMM(count);
220 	for (mmio = gvt->engine_mmio_list.mmio;
221 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
222 		if (mmio->id != ring_id || !mmio->in_context)
223 			continue;
224 
225 		*cs++ = i915_mmio_reg_offset(mmio->reg);
226 		*cs++ = vgpu_vreg_t(vgpu, mmio->reg) | (mmio->mask << 16);
227 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
228 			      *(cs-2), *(cs-1), vgpu->id, ring_id);
229 	}
230 
231 	*cs++ = MI_NOOP;
232 	intel_ring_advance(req, cs);
233 
234 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
235 	if (ret)
236 		return ret;
237 
238 	return 0;
239 }
240 
241 static int
242 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
243 					struct i915_request *req)
244 {
245 	unsigned int index;
246 	u32 *cs;
247 
248 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
249 	if (IS_ERR(cs))
250 		return PTR_ERR(cs);
251 
252 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
253 
254 	for (index = 0; index < GEN9_MOCS_SIZE; index++) {
255 		*cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
256 		*cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
257 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
258 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
259 
260 	}
261 
262 	*cs++ = MI_NOOP;
263 	intel_ring_advance(req, cs);
264 
265 	return 0;
266 }
267 
268 static int
269 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
270 				     struct i915_request *req)
271 {
272 	unsigned int index;
273 	u32 *cs;
274 
275 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
276 	if (IS_ERR(cs))
277 		return PTR_ERR(cs);
278 
279 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
280 
281 	for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
282 		*cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
283 		*cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
284 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
285 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
286 
287 	}
288 
289 	*cs++ = MI_NOOP;
290 	intel_ring_advance(req, cs);
291 
292 	return 0;
293 }
294 
295 /*
296  * Use lri command to initialize the mmio which is in context state image for
297  * inhibit context, it contains tracked engine mmio, render_mocs and
298  * render_mocs_l3cc.
299  */
300 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
301 				       struct i915_request *req)
302 {
303 	int ret;
304 	u32 *cs;
305 
306 	cs = intel_ring_begin(req, 2);
307 	if (IS_ERR(cs))
308 		return PTR_ERR(cs);
309 
310 	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
311 	*cs++ = MI_NOOP;
312 	intel_ring_advance(req, cs);
313 
314 	ret = restore_context_mmio_for_inhibit(vgpu, req);
315 	if (ret)
316 		goto out;
317 
318 	/* no MOCS register in context except render engine */
319 	if (req->engine->id != RCS0)
320 		goto out;
321 
322 	ret = restore_render_mocs_control_for_inhibit(vgpu, req);
323 	if (ret)
324 		goto out;
325 
326 	ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
327 	if (ret)
328 		goto out;
329 
330 out:
331 	cs = intel_ring_begin(req, 2);
332 	if (IS_ERR(cs))
333 		return PTR_ERR(cs);
334 
335 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
336 	*cs++ = MI_NOOP;
337 	intel_ring_advance(req, cs);
338 
339 	return ret;
340 }
341 
342 static u32 gen8_tlb_mmio_offset_list[] = {
343 	[RCS0]  = 0x4260,
344 	[VCS0]  = 0x4264,
345 	[VCS1]  = 0x4268,
346 	[BCS0]  = 0x426c,
347 	[VECS0] = 0x4270,
348 };
349 
350 static void handle_tlb_pending_event(struct intel_vgpu *vgpu,
351 				     const struct intel_engine_cs *engine)
352 {
353 	struct intel_uncore *uncore = engine->uncore;
354 	struct intel_vgpu_submission *s = &vgpu->submission;
355 	u32 *regs = vgpu->gvt->engine_mmio_list.tlb_mmio_offset_list;
356 	u32 cnt = vgpu->gvt->engine_mmio_list.tlb_mmio_offset_list_cnt;
357 	enum forcewake_domains fw;
358 	i915_reg_t reg;
359 
360 	if (!regs)
361 		return;
362 
363 	if (drm_WARN_ON(&engine->i915->drm, engine->id >= cnt))
364 		return;
365 
366 	if (!test_and_clear_bit(engine->id, (void *)s->tlb_handle_pending))
367 		return;
368 
369 	reg = _MMIO(regs[engine->id]);
370 
371 	/* WaForceWakeRenderDuringMmioTLBInvalidate:skl
372 	 * we need to put a forcewake when invalidating RCS TLB caches,
373 	 * otherwise device can go to RC6 state and interrupt invalidation
374 	 * process
375 	 */
376 	fw = intel_uncore_forcewake_for_reg(uncore, reg,
377 					    FW_REG_READ | FW_REG_WRITE);
378 	if (engine->id == RCS0 && GRAPHICS_VER(engine->i915) >= 9)
379 		fw |= FORCEWAKE_RENDER;
380 
381 	intel_uncore_forcewake_get(uncore, fw);
382 
383 	intel_uncore_write_fw(uncore, reg, 0x1);
384 
385 	if (wait_for_atomic(intel_uncore_read_fw(uncore, reg) == 0, 50))
386 		gvt_vgpu_err("timeout in invalidate ring %s tlb\n",
387 			     engine->name);
388 	else
389 		vgpu_vreg_t(vgpu, reg) = 0;
390 
391 	intel_uncore_forcewake_put(uncore, fw);
392 
393 	gvt_dbg_core("invalidate TLB for ring %s\n", engine->name);
394 }
395 
396 static void switch_mocs(struct intel_vgpu *pre, struct intel_vgpu *next,
397 			const struct intel_engine_cs *engine)
398 {
399 	u32 regs[] = {
400 		[RCS0]  = 0xc800,
401 		[VCS0]  = 0xc900,
402 		[VCS1]  = 0xca00,
403 		[BCS0]  = 0xcc00,
404 		[VECS0] = 0xcb00,
405 	};
406 	struct intel_uncore *uncore = engine->uncore;
407 	i915_reg_t offset, l3_offset;
408 	u32 old_v, new_v;
409 	int i;
410 
411 	if (drm_WARN_ON(&engine->i915->drm, engine->id >= ARRAY_SIZE(regs)))
412 		return;
413 
414 	if (engine->id == RCS0 && GRAPHICS_VER(engine->i915) == 9)
415 		return;
416 
417 	if (!pre && !gen9_render_mocs.initialized)
418 		load_render_mocs(engine);
419 
420 	offset.reg = regs[engine->id];
421 	for (i = 0; i < GEN9_MOCS_SIZE; i++) {
422 		if (pre)
423 			old_v = vgpu_vreg_t(pre, offset);
424 		else
425 			old_v = gen9_render_mocs.control_table[engine->id][i];
426 		if (next)
427 			new_v = vgpu_vreg_t(next, offset);
428 		else
429 			new_v = gen9_render_mocs.control_table[engine->id][i];
430 
431 		if (old_v != new_v)
432 			intel_uncore_write_fw(uncore, offset, new_v);
433 
434 		offset.reg += 4;
435 	}
436 
437 	if (engine->id == RCS0) {
438 		l3_offset.reg = 0xb020;
439 		for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
440 			if (pre)
441 				old_v = vgpu_vreg_t(pre, l3_offset);
442 			else
443 				old_v = gen9_render_mocs.l3cc_table[i];
444 			if (next)
445 				new_v = vgpu_vreg_t(next, l3_offset);
446 			else
447 				new_v = gen9_render_mocs.l3cc_table[i];
448 
449 			if (old_v != new_v)
450 				intel_uncore_write_fw(uncore, l3_offset, new_v);
451 
452 			l3_offset.reg += 4;
453 		}
454 	}
455 }
456 
457 #define CTX_CONTEXT_CONTROL_VAL	0x03
458 
459 bool is_inhibit_context(struct intel_context *ce)
460 {
461 	const u32 *reg_state = ce->lrc_reg_state;
462 	u32 inhibit_mask =
463 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
464 
465 	return inhibit_mask ==
466 		(reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
467 }
468 
469 /* Switch ring mmio values (context). */
470 static void switch_mmio(struct intel_vgpu *pre,
471 			struct intel_vgpu *next,
472 			const struct intel_engine_cs *engine)
473 {
474 	struct intel_uncore *uncore = engine->uncore;
475 	struct intel_vgpu_submission *s;
476 	struct engine_mmio *mmio;
477 	u32 old_v, new_v;
478 
479 	if (GRAPHICS_VER(engine->i915) >= 9)
480 		switch_mocs(pre, next, engine);
481 
482 	for (mmio = engine->i915->gvt->engine_mmio_list.mmio;
483 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
484 		if (mmio->id != engine->id)
485 			continue;
486 		/*
487 		 * No need to do save or restore of the mmio which is in context
488 		 * state image on gen9, it's initialized by lri command and
489 		 * save or restore with context together.
490 		 */
491 		if (GRAPHICS_VER(engine->i915) == 9 && mmio->in_context)
492 			continue;
493 
494 		// save
495 		if (pre) {
496 			vgpu_vreg_t(pre, mmio->reg) =
497 				intel_uncore_read_fw(uncore, mmio->reg);
498 			if (mmio->mask)
499 				vgpu_vreg_t(pre, mmio->reg) &=
500 					~(mmio->mask << 16);
501 			old_v = vgpu_vreg_t(pre, mmio->reg);
502 		} else {
503 			old_v = mmio->value =
504 				intel_uncore_read_fw(uncore, mmio->reg);
505 		}
506 
507 		// restore
508 		if (next) {
509 			s = &next->submission;
510 			/*
511 			 * No need to restore the mmio which is in context state
512 			 * image if it's not inhibit context, it will restore
513 			 * itself.
514 			 */
515 			if (mmio->in_context &&
516 			    !is_inhibit_context(s->shadow[engine->id]))
517 				continue;
518 
519 			if (mmio->mask)
520 				new_v = vgpu_vreg_t(next, mmio->reg) |
521 					(mmio->mask << 16);
522 			else
523 				new_v = vgpu_vreg_t(next, mmio->reg);
524 		} else {
525 			if (mmio->in_context)
526 				continue;
527 			if (mmio->mask)
528 				new_v = mmio->value | (mmio->mask << 16);
529 			else
530 				new_v = mmio->value;
531 		}
532 
533 		intel_uncore_write_fw(uncore, mmio->reg, new_v);
534 
535 		trace_render_mmio(pre ? pre->id : 0,
536 				  next ? next->id : 0,
537 				  "switch",
538 				  i915_mmio_reg_offset(mmio->reg),
539 				  old_v, new_v);
540 	}
541 
542 	if (next)
543 		handle_tlb_pending_event(next, engine);
544 }
545 
546 /**
547  * intel_gvt_switch_render_mmio - switch mmio context of specific engine
548  * @pre: the last vGPU that own the engine
549  * @next: the vGPU to switch to
550  * @engine: the engine
551  *
552  * If pre is null indicates that host own the engine. If next is null
553  * indicates that we are switching to host workload.
554  */
555 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
556 			   struct intel_vgpu *next,
557 			   const struct intel_engine_cs *engine)
558 {
559 	if (WARN(!pre && !next, "switch ring %s from host to HOST\n",
560 		 engine->name))
561 		return;
562 
563 	gvt_dbg_render("switch ring %s from %s to %s\n", engine->name,
564 		       pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
565 
566 	/**
567 	 * We are using raw mmio access wrapper to improve the
568 	 * performace for batch mmio read/write, so we need
569 	 * handle forcewake mannually.
570 	 */
571 	intel_uncore_forcewake_get(engine->uncore, FORCEWAKE_ALL);
572 	switch_mmio(pre, next, engine);
573 	intel_uncore_forcewake_put(engine->uncore, FORCEWAKE_ALL);
574 }
575 
576 /**
577  * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list
578  * @gvt: GVT device
579  *
580  */
581 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
582 {
583 	struct engine_mmio *mmio;
584 
585 	if (GRAPHICS_VER(gvt->gt->i915) >= 9) {
586 		gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
587 		gvt->engine_mmio_list.tlb_mmio_offset_list = gen8_tlb_mmio_offset_list;
588 		gvt->engine_mmio_list.tlb_mmio_offset_list_cnt = ARRAY_SIZE(gen8_tlb_mmio_offset_list);
589 		gvt->engine_mmio_list.mocs_mmio_offset_list = gen9_mocs_mmio_offset_list;
590 		gvt->engine_mmio_list.mocs_mmio_offset_list_cnt = ARRAY_SIZE(gen9_mocs_mmio_offset_list);
591 	} else {
592 		gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
593 		gvt->engine_mmio_list.tlb_mmio_offset_list = gen8_tlb_mmio_offset_list;
594 		gvt->engine_mmio_list.tlb_mmio_offset_list_cnt = ARRAY_SIZE(gen8_tlb_mmio_offset_list);
595 	}
596 
597 	for (mmio = gvt->engine_mmio_list.mmio;
598 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
599 		if (mmio->in_context) {
600 			gvt->engine_mmio_list.ctx_mmio_count[mmio->id]++;
601 			intel_gvt_mmio_set_sr_in_ctx(gvt, mmio->reg.reg);
602 		}
603 	}
604 }
605