xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c (revision ddc141e5)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
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 shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include "amdgpu.h"
25 #include "nbio/nbio_6_1_offset.h"
26 #include "nbio/nbio_6_1_sh_mask.h"
27 #include "gc/gc_9_0_offset.h"
28 #include "gc/gc_9_0_sh_mask.h"
29 #include "soc15.h"
30 #include "vega10_ih.h"
31 #include "soc15_common.h"
32 #include "mxgpu_ai.h"
33 
34 static void xgpu_ai_mailbox_send_ack(struct amdgpu_device *adev)
35 {
36 	u32 reg;
37 	int timeout = AI_MAILBOX_TIMEDOUT;
38 	u32 mask = REG_FIELD_MASK(BIF_BX_PF0_MAILBOX_CONTROL, RCV_MSG_VALID);
39 
40 	reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
41 					     mmBIF_BX_PF0_MAILBOX_CONTROL));
42 	reg = REG_SET_FIELD(reg, BIF_BX_PF0_MAILBOX_CONTROL, RCV_MSG_ACK, 1);
43 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
44 				       mmBIF_BX_PF0_MAILBOX_CONTROL), reg);
45 
46 	/*Wait for RCV_MSG_VALID to be 0*/
47 	reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
48 					     mmBIF_BX_PF0_MAILBOX_CONTROL));
49 	while (reg & mask) {
50 		if (timeout <= 0) {
51 			pr_err("RCV_MSG_VALID is not cleared\n");
52 			break;
53 		}
54 		mdelay(1);
55 		timeout -=1;
56 
57 		reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
58 						     mmBIF_BX_PF0_MAILBOX_CONTROL));
59 	}
60 }
61 
62 static void xgpu_ai_mailbox_set_valid(struct amdgpu_device *adev, bool val)
63 {
64 	u32 reg;
65 
66 	reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
67 					     mmBIF_BX_PF0_MAILBOX_CONTROL));
68 	reg = REG_SET_FIELD(reg, BIF_BX_PF0_MAILBOX_CONTROL,
69 			    TRN_MSG_VALID, val ? 1 : 0);
70 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_CONTROL),
71 		      reg);
72 }
73 
74 static int xgpu_ai_mailbox_rcv_msg(struct amdgpu_device *adev,
75 				   enum idh_event event)
76 {
77 	u32 reg;
78 	u32 mask = REG_FIELD_MASK(BIF_BX_PF0_MAILBOX_CONTROL, RCV_MSG_VALID);
79 
80 	if (event != IDH_FLR_NOTIFICATION_CMPL) {
81 		reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
82 						     mmBIF_BX_PF0_MAILBOX_CONTROL));
83 		if (!(reg & mask))
84 			return -ENOENT;
85 	}
86 
87 	reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
88 					     mmBIF_BX_PF0_MAILBOX_MSGBUF_RCV_DW0));
89 	if (reg != event)
90 		return -ENOENT;
91 
92 	xgpu_ai_mailbox_send_ack(adev);
93 
94 	return 0;
95 }
96 
97 static int xgpu_ai_poll_ack(struct amdgpu_device *adev)
98 {
99 	int r = 0, timeout = AI_MAILBOX_TIMEDOUT;
100 	u32 mask = REG_FIELD_MASK(BIF_BX_PF0_MAILBOX_CONTROL, TRN_MSG_ACK);
101 	u32 reg;
102 
103 	reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
104 					     mmBIF_BX_PF0_MAILBOX_CONTROL));
105 	while (!(reg & mask)) {
106 		if (timeout <= 0) {
107 			pr_err("Doesn't get ack from pf.\n");
108 			r = -ETIME;
109 			break;
110 		}
111 		mdelay(5);
112 		timeout -= 5;
113 
114 		reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
115 						     mmBIF_BX_PF0_MAILBOX_CONTROL));
116 	}
117 
118 	return r;
119 }
120 
121 static int xgpu_ai_poll_msg(struct amdgpu_device *adev, enum idh_event event)
122 {
123 	int r = 0, timeout = AI_MAILBOX_TIMEDOUT;
124 
125 	r = xgpu_ai_mailbox_rcv_msg(adev, event);
126 	while (r) {
127 		if (timeout <= 0) {
128 			pr_err("Doesn't get msg:%d from pf.\n", event);
129 			r = -ETIME;
130 			break;
131 		}
132 		mdelay(5);
133 		timeout -= 5;
134 
135 		r = xgpu_ai_mailbox_rcv_msg(adev, event);
136 	}
137 
138 	return r;
139 }
140 
141 static void xgpu_ai_mailbox_trans_msg (struct amdgpu_device *adev,
142 	      enum idh_request req, u32 data1, u32 data2, u32 data3) {
143 	u32 reg;
144 	int r;
145 
146 	reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
147 					     mmBIF_BX_PF0_MAILBOX_MSGBUF_TRN_DW0));
148 	reg = REG_SET_FIELD(reg, BIF_BX_PF0_MAILBOX_MSGBUF_TRN_DW0,
149 			    MSGBUF_DATA, req);
150 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_MSGBUF_TRN_DW0),
151 		      reg);
152 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_MSGBUF_TRN_DW1),
153 				data1);
154 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_MSGBUF_TRN_DW2),
155 				data2);
156 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_MSGBUF_TRN_DW3),
157 				data3);
158 
159 	xgpu_ai_mailbox_set_valid(adev, true);
160 
161 	/* start to poll ack */
162 	r = xgpu_ai_poll_ack(adev);
163 	if (r)
164 		pr_err("Doesn't get ack from pf, continue\n");
165 
166 	xgpu_ai_mailbox_set_valid(adev, false);
167 }
168 
169 static int xgpu_ai_send_access_requests(struct amdgpu_device *adev,
170 					enum idh_request req)
171 {
172 	int r;
173 
174 	xgpu_ai_mailbox_trans_msg(adev, req, 0, 0, 0);
175 
176 	/* start to check msg if request is idh_req_gpu_init_access */
177 	if (req == IDH_REQ_GPU_INIT_ACCESS ||
178 		req == IDH_REQ_GPU_FINI_ACCESS ||
179 		req == IDH_REQ_GPU_RESET_ACCESS) {
180 		r = xgpu_ai_poll_msg(adev, IDH_READY_TO_ACCESS_GPU);
181 		if (r) {
182 			pr_err("Doesn't get READY_TO_ACCESS_GPU from pf, give up\n");
183 			return r;
184 		}
185 		/* Retrieve checksum from mailbox2 */
186 		if (req == IDH_REQ_GPU_INIT_ACCESS) {
187 			adev->virt.fw_reserve.checksum_key =
188 				RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
189 					mmBIF_BX_PF0_MAILBOX_MSGBUF_RCV_DW2));
190 		}
191 	}
192 
193 	return 0;
194 }
195 
196 static int xgpu_ai_request_reset(struct amdgpu_device *adev)
197 {
198 	return xgpu_ai_send_access_requests(adev, IDH_REQ_GPU_RESET_ACCESS);
199 }
200 
201 static int xgpu_ai_request_full_gpu_access(struct amdgpu_device *adev,
202 					   bool init)
203 {
204 	enum idh_request req;
205 
206 	req = init ? IDH_REQ_GPU_INIT_ACCESS : IDH_REQ_GPU_FINI_ACCESS;
207 	return xgpu_ai_send_access_requests(adev, req);
208 }
209 
210 static int xgpu_ai_release_full_gpu_access(struct amdgpu_device *adev,
211 					   bool init)
212 {
213 	enum idh_request req;
214 	int r = 0;
215 
216 	req = init ? IDH_REL_GPU_INIT_ACCESS : IDH_REL_GPU_FINI_ACCESS;
217 	r = xgpu_ai_send_access_requests(adev, req);
218 
219 	return r;
220 }
221 
222 static int xgpu_ai_mailbox_ack_irq(struct amdgpu_device *adev,
223 					struct amdgpu_irq_src *source,
224 					struct amdgpu_iv_entry *entry)
225 {
226 	DRM_DEBUG("get ack intr and do nothing.\n");
227 	return 0;
228 }
229 
230 static int xgpu_ai_set_mailbox_ack_irq(struct amdgpu_device *adev,
231 					struct amdgpu_irq_src *source,
232 					unsigned type,
233 					enum amdgpu_interrupt_state state)
234 {
235 	u32 tmp = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_INT_CNTL));
236 
237 	tmp = REG_SET_FIELD(tmp, BIF_BX_PF0_MAILBOX_INT_CNTL, ACK_INT_EN,
238 				(state == AMDGPU_IRQ_STATE_ENABLE) ? 1 : 0);
239 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_INT_CNTL), tmp);
240 
241 	return 0;
242 }
243 
244 static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
245 {
246 	struct amdgpu_virt *virt = container_of(work, struct amdgpu_virt, flr_work);
247 	struct amdgpu_device *adev = container_of(virt, struct amdgpu_device, virt);
248 
249 	/* wait until RCV_MSG become 3 */
250 	if (xgpu_ai_poll_msg(adev, IDH_FLR_NOTIFICATION_CMPL)) {
251 		pr_err("failed to recieve FLR_CMPL\n");
252 		return;
253 	}
254 
255 	/* Trigger recovery due to world switch failure */
256 	amdgpu_device_gpu_recover(adev, NULL, false);
257 }
258 
259 static int xgpu_ai_set_mailbox_rcv_irq(struct amdgpu_device *adev,
260 				       struct amdgpu_irq_src *src,
261 				       unsigned type,
262 				       enum amdgpu_interrupt_state state)
263 {
264 	u32 tmp = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_INT_CNTL));
265 
266 	tmp = REG_SET_FIELD(tmp, BIF_BX_PF0_MAILBOX_INT_CNTL, VALID_INT_EN,
267 			    (state == AMDGPU_IRQ_STATE_ENABLE) ? 1 : 0);
268 	WREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0, mmBIF_BX_PF0_MAILBOX_INT_CNTL), tmp);
269 
270 	return 0;
271 }
272 
273 static int xgpu_ai_mailbox_rcv_irq(struct amdgpu_device *adev,
274 				   struct amdgpu_irq_src *source,
275 				   struct amdgpu_iv_entry *entry)
276 {
277 	int r;
278 
279 	/* trigger gpu-reset by hypervisor only if TDR disbaled */
280 	if (!amdgpu_gpu_recovery) {
281 		/* see what event we get */
282 		r = xgpu_ai_mailbox_rcv_msg(adev, IDH_FLR_NOTIFICATION);
283 
284 		/* sometimes the interrupt is delayed to inject to VM, so under such case
285 		 * the IDH_FLR_NOTIFICATION is overwritten by VF FLR from GIM side, thus
286 		 * above recieve message could be failed, we should schedule the flr_work
287 		 * anyway
288 		 */
289 		if (r) {
290 			DRM_ERROR("FLR_NOTIFICATION is missed\n");
291 			xgpu_ai_mailbox_send_ack(adev);
292 		}
293 
294 		schedule_work(&adev->virt.flr_work);
295 	}
296 
297 	return 0;
298 }
299 
300 static const struct amdgpu_irq_src_funcs xgpu_ai_mailbox_ack_irq_funcs = {
301 	.set = xgpu_ai_set_mailbox_ack_irq,
302 	.process = xgpu_ai_mailbox_ack_irq,
303 };
304 
305 static const struct amdgpu_irq_src_funcs xgpu_ai_mailbox_rcv_irq_funcs = {
306 	.set = xgpu_ai_set_mailbox_rcv_irq,
307 	.process = xgpu_ai_mailbox_rcv_irq,
308 };
309 
310 void xgpu_ai_mailbox_set_irq_funcs(struct amdgpu_device *adev)
311 {
312 	adev->virt.ack_irq.num_types = 1;
313 	adev->virt.ack_irq.funcs = &xgpu_ai_mailbox_ack_irq_funcs;
314 	adev->virt.rcv_irq.num_types = 1;
315 	adev->virt.rcv_irq.funcs = &xgpu_ai_mailbox_rcv_irq_funcs;
316 }
317 
318 int xgpu_ai_mailbox_add_irq_id(struct amdgpu_device *adev)
319 {
320 	int r;
321 
322 	r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_BIF, 135, &adev->virt.rcv_irq);
323 	if (r)
324 		return r;
325 
326 	r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_BIF, 138, &adev->virt.ack_irq);
327 	if (r) {
328 		amdgpu_irq_put(adev, &adev->virt.rcv_irq, 0);
329 		return r;
330 	}
331 
332 	return 0;
333 }
334 
335 int xgpu_ai_mailbox_get_irq(struct amdgpu_device *adev)
336 {
337 	int r;
338 
339 	r = amdgpu_irq_get(adev, &adev->virt.rcv_irq, 0);
340 	if (r)
341 		return r;
342 	r = amdgpu_irq_get(adev, &adev->virt.ack_irq, 0);
343 	if (r) {
344 		amdgpu_irq_put(adev, &adev->virt.rcv_irq, 0);
345 		return r;
346 	}
347 
348 	INIT_WORK(&adev->virt.flr_work, xgpu_ai_mailbox_flr_work);
349 
350 	return 0;
351 }
352 
353 void xgpu_ai_mailbox_put_irq(struct amdgpu_device *adev)
354 {
355 	amdgpu_irq_put(adev, &adev->virt.ack_irq, 0);
356 	amdgpu_irq_put(adev, &adev->virt.rcv_irq, 0);
357 }
358 
359 const struct amdgpu_virt_ops xgpu_ai_virt_ops = {
360 	.req_full_gpu	= xgpu_ai_request_full_gpu_access,
361 	.rel_full_gpu	= xgpu_ai_release_full_gpu_access,
362 	.reset_gpu = xgpu_ai_request_reset,
363 	.wait_reset = NULL,
364 	.trans_msg = xgpu_ai_mailbox_trans_msg,
365 };
366