xref: /linux/drivers/gpu/drm/i915/gvt/mmio_context.c (revision d30c1683aaecb93d2ab95685dc4300a33d3cea7a)
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 <drm/drm_print.h>
37 
38 #include "gt/intel_context.h"
39 #include "gt/intel_engine_regs.h"
40 #include "gt/intel_gpu_commands.h"
41 #include "gt/intel_gt_regs.h"
42 #include "gt/intel_ring.h"
43 
44 #include "gvt.h"
45 #include "i915_drv.h"
46 #include "i915_reg.h"
47 #include "i915_wait_util.h"
48 #include "trace.h"
49 
50 #define GEN9_MOCS_SIZE		64
51 
52 struct engine_mmio {
53 	enum intel_engine_id id;
54 	i915_reg_t reg;
55 	u32 mask;
56 	bool in_context;
57 	u32 value;
58 };
59 
60 /* Raw offset is append to each line for convenience. */
61 static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = {
62 	{RCS0, RING_MODE_GEN7(RENDER_RING_BASE), 0xffff, false}, /* 0x229c */
63 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
64 	{RCS0, HWSTAM, 0x0, false}, /* 0x2098 */
65 	{RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */
66 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
67 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
68 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
69 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
70 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
71 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
72 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
73 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
74 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
75 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
76 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
77 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
78 	{RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
79 	{RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
80 	{RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
81 	{RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
82 	{RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
83 	{RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
84 
85 	{BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
86 	{BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
87 	{BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
88 	{BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
89 	{BCS0, RING_EXCC(BLT_RING_BASE), 0xffff, false}, /* 0x22028 */
90 	{RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */
91 };
92 
93 static struct engine_mmio gen9_engine_mmio_list[] __cacheline_aligned = {
94 	{RCS0, RING_MODE_GEN7(RENDER_RING_BASE), 0xffff, false}, /* 0x229c */
95 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
96 	{RCS0, HWSTAM, 0x0, false}, /* 0x2098 */
97 	{RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */
98 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */
99 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */
100 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */
101 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */
102 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */
103 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */
104 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */
105 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */
106 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */
107 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */
108 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */
109 	{RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */
110 	{RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */
111 	{RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */
112 	{RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */
113 	{RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */
114 	{RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */
115 	{RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */
116 
117 	{RCS0, GEN8_PRIVATE_PAT_LO, 0, false}, /* 0x40e0 */
118 	{RCS0, GEN8_PRIVATE_PAT_HI, 0, false}, /* 0x40e4 */
119 	{RCS0, GEN8_CS_CHICKEN1, 0xffff, true}, /* 0x2580 */
120 	{RCS0, COMMON_SLICE_CHICKEN2, 0xffff, true}, /* 0x7014 */
121 	{RCS0, GEN9_CS_DEBUG_MODE1, 0xffff, false}, /* 0x20ec */
122 	{RCS0, _MMIO(0xb118), 0, false}, /* GEN8_L3SQCREG4 */
123 	{RCS0, _MMIO(0xb11c), 0, false}, /* GEN9_SCRATCH1 */
124 	{RCS0, GEN9_SCRATCH_LNCF1, 0, false}, /* 0xb008 */
125 	{RCS0, GEN7_HALF_SLICE_CHICKEN1, 0xffff, true}, /* 0xe100 */
126 	{RCS0, _MMIO(0xe180), 0xffff, true}, /* HALF_SLICE_CHICKEN2 */
127 	{RCS0, _MMIO(0xe184), 0xffff, true}, /* GEN8_HALF_SLICE_CHICKEN3 */
128 	{RCS0, _MMIO(0xe188), 0xffff, true}, /* GEN9_HALF_SLICE_CHICKEN5 */
129 	{RCS0, _MMIO(0xe194), 0xffff, true}, /* GEN9_HALF_SLICE_CHICKEN7 */
130 	{RCS0, _MMIO(0xe4f0), 0xffff, true}, /* GEN8_ROW_CHICKEN */
131 	{RCS0, TRVATTL3PTRDW(0), 0, true}, /* 0x4de0 */
132 	{RCS0, TRVATTL3PTRDW(1), 0, true}, /* 0x4de4 */
133 	{RCS0, TRNULLDETCT, 0, true}, /* 0x4de8 */
134 	{RCS0, TRINVTILEDETCT, 0, true}, /* 0x4dec */
135 	{RCS0, TRVADR, 0, true}, /* 0x4df0 */
136 	{RCS0, TRTTE, 0, true}, /* 0x4df4 */
137 	{RCS0, _MMIO(0x4dfc), 0, true},
138 
139 	{BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */
140 	{BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */
141 	{BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */
142 	{BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */
143 	{BCS0, RING_EXCC(BLT_RING_BASE), 0xffff, false}, /* 0x22028 */
144 
145 	{VCS1, RING_EXCC(GEN8_BSD2_RING_BASE), 0xffff, false}, /* 0x1c028 */
146 
147 	{VECS0, RING_EXCC(VEBOX_RING_BASE), 0xffff, false}, /* 0x1a028 */
148 
149 	{RCS0, GEN8_HDC_CHICKEN1, 0xffff, true}, /* 0x7304 */
150 	{RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */
151 	{RCS0, GEN7_UCGCTL4, 0x0, false}, /* 0x940c */
152 	{RCS0, GAMT_CHKN_BIT_REG, 0x0, false}, /* 0x4ab8 */
153 
154 	{RCS0, GEN9_GAMT_ECO_REG_RW_IA, 0x0, false}, /* 0x4ab0 */
155 	{RCS0, GEN9_CSFE_CHICKEN1_RCS, 0xffff, false}, /* 0x20d4 */
156 	{RCS0, _MMIO(0x20D8), 0xffff, true}, /* 0x20d8 */
157 
158 	{RCS0, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */
159 	{RCS0, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */
160 	{RCS0, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */
161 	{RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */
162 };
163 
164 static struct {
165 	bool initialized;
166 	u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
167 	u32 l3cc_table[GEN9_MOCS_SIZE / 2];
168 } gen9_render_mocs;
169 
170 static u32 gen9_mocs_mmio_offset_list[] = {
171 	[RCS0]  = 0xc800,
172 	[VCS0]  = 0xc900,
173 	[VCS1]  = 0xca00,
174 	[BCS0]  = 0xcc00,
175 	[VECS0] = 0xcb00,
176 };
177 
178 static void load_render_mocs(const struct intel_engine_cs *engine)
179 {
180 	struct intel_gvt *gvt = engine->i915->gvt;
181 	struct intel_uncore *uncore = engine->uncore;
182 	u32 cnt = gvt->engine_mmio_list.mocs_mmio_offset_list_cnt;
183 	u32 *regs = gvt->engine_mmio_list.mocs_mmio_offset_list;
184 	i915_reg_t offset;
185 	int ring_id, i;
186 
187 	/* Platform doesn't have mocs mmios. */
188 	if (!regs)
189 		return;
190 
191 	for (ring_id = 0; ring_id < cnt; ring_id++) {
192 		if (!HAS_ENGINE(engine->gt, ring_id))
193 			continue;
194 
195 		offset.reg = regs[ring_id];
196 		for (i = 0; i < GEN9_MOCS_SIZE; i++) {
197 			gen9_render_mocs.control_table[ring_id][i] =
198 				intel_uncore_read_fw(uncore, offset);
199 			offset.reg += 4;
200 		}
201 	}
202 
203 	offset.reg = 0xb020;
204 	for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
205 		gen9_render_mocs.l3cc_table[i] =
206 			intel_uncore_read_fw(uncore, offset);
207 		offset.reg += 4;
208 	}
209 	gen9_render_mocs.initialized = true;
210 }
211 
212 static int
213 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
214 				 struct i915_request *req)
215 {
216 	u32 *cs;
217 	int ret;
218 	struct engine_mmio *mmio;
219 	struct intel_gvt *gvt = vgpu->gvt;
220 	int ring_id = req->engine->id;
221 	int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
222 
223 	if (count == 0)
224 		return 0;
225 
226 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
227 	if (ret)
228 		return ret;
229 
230 	cs = intel_ring_begin(req, count * 2 + 2);
231 	if (IS_ERR(cs))
232 		return PTR_ERR(cs);
233 
234 	*cs++ = MI_LOAD_REGISTER_IMM(count);
235 	for (mmio = gvt->engine_mmio_list.mmio;
236 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
237 		if (mmio->id != ring_id || !mmio->in_context)
238 			continue;
239 
240 		*cs++ = i915_mmio_reg_offset(mmio->reg);
241 		*cs++ = vgpu_vreg_t(vgpu, mmio->reg) | (mmio->mask << 16);
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, ring_id);
244 	}
245 
246 	*cs++ = MI_NOOP;
247 	intel_ring_advance(req, cs);
248 
249 	ret = req->engine->emit_flush(req, EMIT_BARRIER);
250 	if (ret)
251 		return ret;
252 
253 	return 0;
254 }
255 
256 static int
257 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
258 					struct i915_request *req)
259 {
260 	unsigned int index;
261 	u32 *cs;
262 
263 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
264 	if (IS_ERR(cs))
265 		return PTR_ERR(cs);
266 
267 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
268 
269 	for (index = 0; index < GEN9_MOCS_SIZE; index++) {
270 		*cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
271 		*cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
272 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
273 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
274 
275 	}
276 
277 	*cs++ = MI_NOOP;
278 	intel_ring_advance(req, cs);
279 
280 	return 0;
281 }
282 
283 static int
284 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
285 				     struct i915_request *req)
286 {
287 	unsigned int index;
288 	u32 *cs;
289 
290 	cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
291 	if (IS_ERR(cs))
292 		return PTR_ERR(cs);
293 
294 	*cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
295 
296 	for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
297 		*cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
298 		*cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
299 		gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
300 			      *(cs-2), *(cs-1), vgpu->id, req->engine->id);
301 
302 	}
303 
304 	*cs++ = MI_NOOP;
305 	intel_ring_advance(req, cs);
306 
307 	return 0;
308 }
309 
310 /*
311  * Use lri command to initialize the mmio which is in context state image for
312  * inhibit context, it contains tracked engine mmio, render_mocs and
313  * render_mocs_l3cc.
314  */
315 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
316 				       struct i915_request *req)
317 {
318 	int ret;
319 	u32 *cs;
320 
321 	cs = intel_ring_begin(req, 2);
322 	if (IS_ERR(cs))
323 		return PTR_ERR(cs);
324 
325 	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
326 	*cs++ = MI_NOOP;
327 	intel_ring_advance(req, cs);
328 
329 	ret = restore_context_mmio_for_inhibit(vgpu, req);
330 	if (ret)
331 		goto out;
332 
333 	/* no MOCS register in context except render engine */
334 	if (req->engine->id != RCS0)
335 		goto out;
336 
337 	ret = restore_render_mocs_control_for_inhibit(vgpu, req);
338 	if (ret)
339 		goto out;
340 
341 	ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
342 	if (ret)
343 		goto out;
344 
345 out:
346 	cs = intel_ring_begin(req, 2);
347 	if (IS_ERR(cs))
348 		return PTR_ERR(cs);
349 
350 	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
351 	*cs++ = MI_NOOP;
352 	intel_ring_advance(req, cs);
353 
354 	return ret;
355 }
356 
357 static u32 gen8_tlb_mmio_offset_list[] = {
358 	[RCS0]  = 0x4260,
359 	[VCS0]  = 0x4264,
360 	[VCS1]  = 0x4268,
361 	[BCS0]  = 0x426c,
362 	[VECS0] = 0x4270,
363 };
364 
365 static void handle_tlb_pending_event(struct intel_vgpu *vgpu,
366 				     const struct intel_engine_cs *engine)
367 {
368 	struct intel_uncore *uncore = engine->uncore;
369 	struct intel_vgpu_submission *s = &vgpu->submission;
370 	u32 *regs = vgpu->gvt->engine_mmio_list.tlb_mmio_offset_list;
371 	u32 cnt = vgpu->gvt->engine_mmio_list.tlb_mmio_offset_list_cnt;
372 	enum forcewake_domains fw;
373 	i915_reg_t reg;
374 
375 	if (!regs)
376 		return;
377 
378 	if (drm_WARN_ON(&engine->i915->drm, engine->id >= cnt))
379 		return;
380 
381 	if (!test_and_clear_bit(engine->id, (void *)s->tlb_handle_pending))
382 		return;
383 
384 	reg = _MMIO(regs[engine->id]);
385 
386 	/* WaForceWakeRenderDuringMmioTLBInvalidate:skl
387 	 * we need to put a forcewake when invalidating RCS TLB caches,
388 	 * otherwise device can go to RC6 state and interrupt invalidation
389 	 * process
390 	 */
391 	fw = intel_uncore_forcewake_for_reg(uncore, reg,
392 					    FW_REG_READ | FW_REG_WRITE);
393 	if (engine->id == RCS0 && GRAPHICS_VER(engine->i915) >= 9)
394 		fw |= FORCEWAKE_RENDER;
395 
396 	intel_uncore_forcewake_get(uncore, fw);
397 
398 	intel_uncore_write_fw(uncore, reg, 0x1);
399 
400 	if (wait_for_atomic(intel_uncore_read_fw(uncore, reg) == 0, 50))
401 		gvt_vgpu_err("timeout in invalidate ring %s tlb\n",
402 			     engine->name);
403 	else
404 		vgpu_vreg_t(vgpu, reg) = 0;
405 
406 	intel_uncore_forcewake_put(uncore, fw);
407 
408 	gvt_dbg_core("invalidate TLB for ring %s\n", engine->name);
409 }
410 
411 static void switch_mocs(struct intel_vgpu *pre, struct intel_vgpu *next,
412 			const struct intel_engine_cs *engine)
413 {
414 	u32 regs[] = {
415 		[RCS0]  = 0xc800,
416 		[VCS0]  = 0xc900,
417 		[VCS1]  = 0xca00,
418 		[BCS0]  = 0xcc00,
419 		[VECS0] = 0xcb00,
420 	};
421 	struct intel_uncore *uncore = engine->uncore;
422 	i915_reg_t offset, l3_offset;
423 	u32 old_v, new_v;
424 	int i;
425 
426 	if (drm_WARN_ON(&engine->i915->drm, engine->id >= ARRAY_SIZE(regs)))
427 		return;
428 
429 	if (engine->id == RCS0 && GRAPHICS_VER(engine->i915) == 9)
430 		return;
431 
432 	if (!pre && !gen9_render_mocs.initialized)
433 		load_render_mocs(engine);
434 
435 	offset.reg = regs[engine->id];
436 	for (i = 0; i < GEN9_MOCS_SIZE; i++) {
437 		if (pre)
438 			old_v = vgpu_vreg_t(pre, offset);
439 		else
440 			old_v = gen9_render_mocs.control_table[engine->id][i];
441 		if (next)
442 			new_v = vgpu_vreg_t(next, offset);
443 		else
444 			new_v = gen9_render_mocs.control_table[engine->id][i];
445 
446 		if (old_v != new_v)
447 			intel_uncore_write_fw(uncore, offset, new_v);
448 
449 		offset.reg += 4;
450 	}
451 
452 	if (engine->id == RCS0) {
453 		l3_offset.reg = 0xb020;
454 		for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
455 			if (pre)
456 				old_v = vgpu_vreg_t(pre, l3_offset);
457 			else
458 				old_v = gen9_render_mocs.l3cc_table[i];
459 			if (next)
460 				new_v = vgpu_vreg_t(next, l3_offset);
461 			else
462 				new_v = gen9_render_mocs.l3cc_table[i];
463 
464 			if (old_v != new_v)
465 				intel_uncore_write_fw(uncore, l3_offset, new_v);
466 
467 			l3_offset.reg += 4;
468 		}
469 	}
470 }
471 
472 #define CTX_CONTEXT_CONTROL_VAL	0x03
473 
474 bool is_inhibit_context(struct intel_context *ce)
475 {
476 	const u32 *reg_state = ce->lrc_reg_state;
477 	u32 inhibit_mask =
478 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
479 
480 	return inhibit_mask ==
481 		(reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
482 }
483 
484 /* Switch ring mmio values (context). */
485 static void switch_mmio(struct intel_vgpu *pre,
486 			struct intel_vgpu *next,
487 			const struct intel_engine_cs *engine)
488 {
489 	struct intel_uncore *uncore = engine->uncore;
490 	struct intel_vgpu_submission *s;
491 	struct engine_mmio *mmio;
492 	u32 old_v, new_v;
493 
494 	if (GRAPHICS_VER(engine->i915) >= 9)
495 		switch_mocs(pre, next, engine);
496 
497 	for (mmio = engine->i915->gvt->engine_mmio_list.mmio;
498 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
499 		if (mmio->id != engine->id)
500 			continue;
501 		/*
502 		 * No need to do save or restore of the mmio which is in context
503 		 * state image on gen9, it's initialized by lri command and
504 		 * save or restore with context together.
505 		 */
506 		if (GRAPHICS_VER(engine->i915) == 9 && mmio->in_context)
507 			continue;
508 
509 		// save
510 		if (pre) {
511 			vgpu_vreg_t(pre, mmio->reg) =
512 				intel_uncore_read_fw(uncore, mmio->reg);
513 			if (mmio->mask)
514 				vgpu_vreg_t(pre, mmio->reg) &=
515 					~(mmio->mask << 16);
516 			old_v = vgpu_vreg_t(pre, mmio->reg);
517 		} else {
518 			old_v = mmio->value =
519 				intel_uncore_read_fw(uncore, mmio->reg);
520 		}
521 
522 		// restore
523 		if (next) {
524 			s = &next->submission;
525 			/*
526 			 * No need to restore the mmio which is in context state
527 			 * image if it's not inhibit context, it will restore
528 			 * itself.
529 			 */
530 			if (mmio->in_context &&
531 			    !is_inhibit_context(s->shadow[engine->id]))
532 				continue;
533 
534 			if (mmio->mask)
535 				new_v = vgpu_vreg_t(next, mmio->reg) |
536 					(mmio->mask << 16);
537 			else
538 				new_v = vgpu_vreg_t(next, mmio->reg);
539 		} else {
540 			if (mmio->in_context)
541 				continue;
542 			if (mmio->mask)
543 				new_v = mmio->value | (mmio->mask << 16);
544 			else
545 				new_v = mmio->value;
546 		}
547 
548 		intel_uncore_write_fw(uncore, mmio->reg, new_v);
549 
550 		trace_render_mmio(pre ? pre->id : 0,
551 				  next ? next->id : 0,
552 				  "switch",
553 				  i915_mmio_reg_offset(mmio->reg),
554 				  old_v, new_v);
555 	}
556 
557 	if (next)
558 		handle_tlb_pending_event(next, engine);
559 }
560 
561 /**
562  * intel_gvt_switch_mmio - switch mmio context of specific engine
563  * @pre: the last vGPU that own the engine
564  * @next: the vGPU to switch to
565  * @engine: the engine
566  *
567  * If pre is null indicates that host own the engine. If next is null
568  * indicates that we are switching to host workload.
569  */
570 void intel_gvt_switch_mmio(struct intel_vgpu *pre,
571 			   struct intel_vgpu *next,
572 			   const struct intel_engine_cs *engine)
573 {
574 	if (WARN(!pre && !next, "switch ring %s from host to HOST\n",
575 		 engine->name))
576 		return;
577 
578 	gvt_dbg_render("switch ring %s from %s to %s\n", engine->name,
579 		       pre ? "vGPU" : "host", next ? "vGPU" : "HOST");
580 
581 	/**
582 	 * We are using raw mmio access wrapper to improve the
583 	 * performance for batch mmio read/write, so we need
584 	 * handle forcewake manually.
585 	 */
586 	intel_uncore_forcewake_get(engine->uncore, FORCEWAKE_ALL);
587 	switch_mmio(pre, next, engine);
588 	intel_uncore_forcewake_put(engine->uncore, FORCEWAKE_ALL);
589 }
590 
591 /**
592  * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list
593  * @gvt: GVT device
594  *
595  */
596 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
597 {
598 	struct engine_mmio *mmio;
599 
600 	if (GRAPHICS_VER(gvt->gt->i915) >= 9) {
601 		gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
602 		gvt->engine_mmio_list.tlb_mmio_offset_list = gen8_tlb_mmio_offset_list;
603 		gvt->engine_mmio_list.tlb_mmio_offset_list_cnt = ARRAY_SIZE(gen8_tlb_mmio_offset_list);
604 		gvt->engine_mmio_list.mocs_mmio_offset_list = gen9_mocs_mmio_offset_list;
605 		gvt->engine_mmio_list.mocs_mmio_offset_list_cnt = ARRAY_SIZE(gen9_mocs_mmio_offset_list);
606 	} else {
607 		gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
608 		gvt->engine_mmio_list.tlb_mmio_offset_list = gen8_tlb_mmio_offset_list;
609 		gvt->engine_mmio_list.tlb_mmio_offset_list_cnt = ARRAY_SIZE(gen8_tlb_mmio_offset_list);
610 	}
611 
612 	for (mmio = gvt->engine_mmio_list.mmio;
613 	     i915_mmio_reg_valid(mmio->reg); mmio++) {
614 		if (mmio->in_context) {
615 			gvt->engine_mmio_list.ctx_mmio_count[mmio->id]++;
616 			intel_gvt_mmio_set_sr_in_ctx(gvt, mmio->reg.reg);
617 		}
618 	}
619 }
620