xref: /openbmc/linux/drivers/accel/ivpu/ivpu_ipc.c (revision 8d81cd1a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2020-2023 Intel Corporation
4  */
5 
6 #include <linux/genalloc.h>
7 #include <linux/highmem.h>
8 #include <linux/kthread.h>
9 #include <linux/wait.h>
10 
11 #include "ivpu_drv.h"
12 #include "ivpu_gem.h"
13 #include "ivpu_hw.h"
14 #include "ivpu_hw_reg_io.h"
15 #include "ivpu_ipc.h"
16 #include "ivpu_jsm_msg.h"
17 #include "ivpu_pm.h"
18 
19 #define IPC_MAX_RX_MSG	128
20 #define IS_KTHREAD()	(get_current()->flags & PF_KTHREAD)
21 
22 struct ivpu_ipc_tx_buf {
23 	struct ivpu_ipc_hdr ipc;
24 	struct vpu_jsm_msg jsm;
25 };
26 
27 struct ivpu_ipc_rx_msg {
28 	struct list_head link;
29 	struct ivpu_ipc_hdr *ipc_hdr;
30 	struct vpu_jsm_msg *jsm_msg;
31 };
32 
33 static void ivpu_ipc_msg_dump(struct ivpu_device *vdev, char *c,
34 			      struct ivpu_ipc_hdr *ipc_hdr, u32 vpu_addr)
35 {
36 	ivpu_dbg(vdev, IPC,
37 		 "%s: vpu:0x%x (data_addr:0x%08x, data_size:0x%x, channel:0x%x, src_node:0x%x, dst_node:0x%x, status:0x%x)",
38 		 c, vpu_addr, ipc_hdr->data_addr, ipc_hdr->data_size, ipc_hdr->channel,
39 		 ipc_hdr->src_node, ipc_hdr->dst_node, ipc_hdr->status);
40 }
41 
42 static void ivpu_jsm_msg_dump(struct ivpu_device *vdev, char *c,
43 			      struct vpu_jsm_msg *jsm_msg, u32 vpu_addr)
44 {
45 	u32 *payload = (u32 *)&jsm_msg->payload;
46 
47 	ivpu_dbg(vdev, JSM,
48 		 "%s: vpu:0x%08x (type:0x%x, status:0x%x, id: 0x%x, result: 0x%x, payload:0x%x 0x%x 0x%x 0x%x 0x%x)\n",
49 		 c, vpu_addr, jsm_msg->type, jsm_msg->status, jsm_msg->request_id, jsm_msg->result,
50 		 payload[0], payload[1], payload[2], payload[3], payload[4]);
51 }
52 
53 static void
54 ivpu_ipc_rx_mark_free(struct ivpu_device *vdev, struct ivpu_ipc_hdr *ipc_hdr,
55 		      struct vpu_jsm_msg *jsm_msg)
56 {
57 	ipc_hdr->status = IVPU_IPC_HDR_FREE;
58 	if (jsm_msg)
59 		jsm_msg->status = VPU_JSM_MSG_FREE;
60 	wmb(); /* Flush WC buffers for message statuses */
61 }
62 
63 static void ivpu_ipc_mem_fini(struct ivpu_device *vdev)
64 {
65 	struct ivpu_ipc_info *ipc = vdev->ipc;
66 
67 	ivpu_bo_free_internal(ipc->mem_rx);
68 	ivpu_bo_free_internal(ipc->mem_tx);
69 }
70 
71 static int
72 ivpu_ipc_tx_prepare(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons,
73 		    struct vpu_jsm_msg *req)
74 {
75 	struct ivpu_ipc_info *ipc = vdev->ipc;
76 	struct ivpu_ipc_tx_buf *tx_buf;
77 	u32 tx_buf_vpu_addr;
78 	u32 jsm_vpu_addr;
79 
80 	tx_buf_vpu_addr = gen_pool_alloc(ipc->mm_tx, sizeof(*tx_buf));
81 	if (!tx_buf_vpu_addr) {
82 		ivpu_err(vdev, "Failed to reserve IPC buffer, size %ld\n",
83 			 sizeof(*tx_buf));
84 		return -ENOMEM;
85 	}
86 
87 	tx_buf = ivpu_to_cpu_addr(ipc->mem_tx, tx_buf_vpu_addr);
88 	if (drm_WARN_ON(&vdev->drm, !tx_buf)) {
89 		gen_pool_free(ipc->mm_tx, tx_buf_vpu_addr, sizeof(*tx_buf));
90 		return -EIO;
91 	}
92 
93 	jsm_vpu_addr = tx_buf_vpu_addr + offsetof(struct ivpu_ipc_tx_buf, jsm);
94 
95 	if (tx_buf->ipc.status != IVPU_IPC_HDR_FREE)
96 		ivpu_warn(vdev, "IPC message vpu:0x%x not released by firmware\n",
97 			  tx_buf_vpu_addr);
98 
99 	if (tx_buf->jsm.status != VPU_JSM_MSG_FREE)
100 		ivpu_warn(vdev, "JSM message vpu:0x%x not released by firmware\n",
101 			  jsm_vpu_addr);
102 
103 	memset(tx_buf, 0, sizeof(*tx_buf));
104 	tx_buf->ipc.data_addr = jsm_vpu_addr;
105 	/* TODO: Set data_size to actual JSM message size, not union of all messages */
106 	tx_buf->ipc.data_size = sizeof(*req);
107 	tx_buf->ipc.channel = cons->channel;
108 	tx_buf->ipc.src_node = 0;
109 	tx_buf->ipc.dst_node = 1;
110 	tx_buf->ipc.status = IVPU_IPC_HDR_ALLOCATED;
111 	tx_buf->jsm.type = req->type;
112 	tx_buf->jsm.status = VPU_JSM_MSG_ALLOCATED;
113 	tx_buf->jsm.payload = req->payload;
114 
115 	req->request_id = atomic_inc_return(&ipc->request_id);
116 	tx_buf->jsm.request_id = req->request_id;
117 	cons->request_id = req->request_id;
118 	wmb(); /* Flush WC buffers for IPC, JSM msgs */
119 
120 	cons->tx_vpu_addr = tx_buf_vpu_addr;
121 
122 	ivpu_jsm_msg_dump(vdev, "TX", &tx_buf->jsm, jsm_vpu_addr);
123 	ivpu_ipc_msg_dump(vdev, "TX", &tx_buf->ipc, tx_buf_vpu_addr);
124 
125 	return 0;
126 }
127 
128 static void ivpu_ipc_tx_release(struct ivpu_device *vdev, u32 vpu_addr)
129 {
130 	struct ivpu_ipc_info *ipc = vdev->ipc;
131 
132 	if (vpu_addr)
133 		gen_pool_free(ipc->mm_tx, vpu_addr, sizeof(struct ivpu_ipc_tx_buf));
134 }
135 
136 static void ivpu_ipc_tx(struct ivpu_device *vdev, u32 vpu_addr)
137 {
138 	ivpu_hw_reg_ipc_tx_set(vdev, vpu_addr);
139 }
140 
141 void
142 ivpu_ipc_consumer_add(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons, u32 channel)
143 {
144 	struct ivpu_ipc_info *ipc = vdev->ipc;
145 
146 	INIT_LIST_HEAD(&cons->link);
147 	cons->channel = channel;
148 	cons->tx_vpu_addr = 0;
149 	cons->request_id = 0;
150 	spin_lock_init(&cons->rx_msg_lock);
151 	INIT_LIST_HEAD(&cons->rx_msg_list);
152 	init_waitqueue_head(&cons->rx_msg_wq);
153 
154 	spin_lock_irq(&ipc->cons_list_lock);
155 	list_add_tail(&cons->link, &ipc->cons_list);
156 	spin_unlock_irq(&ipc->cons_list_lock);
157 }
158 
159 void ivpu_ipc_consumer_del(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons)
160 {
161 	struct ivpu_ipc_info *ipc = vdev->ipc;
162 	struct ivpu_ipc_rx_msg *rx_msg, *r;
163 
164 	spin_lock_irq(&ipc->cons_list_lock);
165 	list_del(&cons->link);
166 	spin_unlock_irq(&ipc->cons_list_lock);
167 
168 	spin_lock_irq(&cons->rx_msg_lock);
169 	list_for_each_entry_safe(rx_msg, r, &cons->rx_msg_list, link) {
170 		list_del(&rx_msg->link);
171 		ivpu_ipc_rx_mark_free(vdev, rx_msg->ipc_hdr, rx_msg->jsm_msg);
172 		atomic_dec(&ipc->rx_msg_count);
173 		kfree(rx_msg);
174 	}
175 	spin_unlock_irq(&cons->rx_msg_lock);
176 
177 	ivpu_ipc_tx_release(vdev, cons->tx_vpu_addr);
178 }
179 
180 static int
181 ivpu_ipc_send(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons, struct vpu_jsm_msg *req)
182 {
183 	struct ivpu_ipc_info *ipc = vdev->ipc;
184 	int ret;
185 
186 	mutex_lock(&ipc->lock);
187 
188 	if (!ipc->on) {
189 		ret = -EAGAIN;
190 		goto unlock;
191 	}
192 
193 	ret = ivpu_ipc_tx_prepare(vdev, cons, req);
194 	if (ret)
195 		goto unlock;
196 
197 	ivpu_ipc_tx(vdev, cons->tx_vpu_addr);
198 
199 unlock:
200 	mutex_unlock(&ipc->lock);
201 	return ret;
202 }
203 
204 int ivpu_ipc_receive(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons,
205 		     struct ivpu_ipc_hdr *ipc_buf,
206 		     struct vpu_jsm_msg *ipc_payload, unsigned long timeout_ms)
207 {
208 	struct ivpu_ipc_info *ipc = vdev->ipc;
209 	struct ivpu_ipc_rx_msg *rx_msg;
210 	int wait_ret, ret = 0;
211 
212 	wait_ret = wait_event_timeout(cons->rx_msg_wq,
213 				      (IS_KTHREAD() && kthread_should_stop()) ||
214 				      !list_empty(&cons->rx_msg_list),
215 				      msecs_to_jiffies(timeout_ms));
216 
217 	if (IS_KTHREAD() && kthread_should_stop())
218 		return -EINTR;
219 
220 	if (wait_ret == 0)
221 		return -ETIMEDOUT;
222 
223 	spin_lock_irq(&cons->rx_msg_lock);
224 	rx_msg = list_first_entry_or_null(&cons->rx_msg_list, struct ivpu_ipc_rx_msg, link);
225 	if (!rx_msg) {
226 		spin_unlock_irq(&cons->rx_msg_lock);
227 		return -EAGAIN;
228 	}
229 	list_del(&rx_msg->link);
230 	spin_unlock_irq(&cons->rx_msg_lock);
231 
232 	if (ipc_buf)
233 		memcpy(ipc_buf, rx_msg->ipc_hdr, sizeof(*ipc_buf));
234 	if (rx_msg->jsm_msg) {
235 		u32 size = min_t(int, rx_msg->ipc_hdr->data_size, sizeof(*ipc_payload));
236 
237 		if (rx_msg->jsm_msg->result != VPU_JSM_STATUS_SUCCESS) {
238 			ivpu_dbg(vdev, IPC, "IPC resp result error: %d\n", rx_msg->jsm_msg->result);
239 			ret = -EBADMSG;
240 		}
241 
242 		if (ipc_payload)
243 			memcpy(ipc_payload, rx_msg->jsm_msg, size);
244 	}
245 
246 	ivpu_ipc_rx_mark_free(vdev, rx_msg->ipc_hdr, rx_msg->jsm_msg);
247 	atomic_dec(&ipc->rx_msg_count);
248 	kfree(rx_msg);
249 
250 	return ret;
251 }
252 
253 static int
254 ivpu_ipc_send_receive_internal(struct ivpu_device *vdev, struct vpu_jsm_msg *req,
255 			       enum vpu_ipc_msg_type expected_resp_type,
256 			       struct vpu_jsm_msg *resp, u32 channel,
257 			       unsigned long timeout_ms)
258 {
259 	struct ivpu_ipc_consumer cons;
260 	int ret;
261 
262 	ivpu_ipc_consumer_add(vdev, &cons, channel);
263 
264 	ret = ivpu_ipc_send(vdev, &cons, req);
265 	if (ret) {
266 		ivpu_warn(vdev, "IPC send failed: %d\n", ret);
267 		goto consumer_del;
268 	}
269 
270 	ret = ivpu_ipc_receive(vdev, &cons, NULL, resp, timeout_ms);
271 	if (ret) {
272 		ivpu_warn(vdev, "IPC receive failed: type 0x%x, ret %d\n", req->type, ret);
273 		goto consumer_del;
274 	}
275 
276 	if (resp->type != expected_resp_type) {
277 		ivpu_warn(vdev, "Invalid JSM response type: 0x%x\n", resp->type);
278 		ret = -EBADE;
279 	}
280 
281 consumer_del:
282 	ivpu_ipc_consumer_del(vdev, &cons);
283 	return ret;
284 }
285 
286 int ivpu_ipc_send_receive(struct ivpu_device *vdev, struct vpu_jsm_msg *req,
287 			  enum vpu_ipc_msg_type expected_resp_type,
288 			  struct vpu_jsm_msg *resp, u32 channel,
289 			  unsigned long timeout_ms)
290 {
291 	struct vpu_jsm_msg hb_req = { .type = VPU_JSM_MSG_QUERY_ENGINE_HB };
292 	struct vpu_jsm_msg hb_resp;
293 	int ret, hb_ret;
294 
295 	ret = ivpu_rpm_get(vdev);
296 	if (ret < 0)
297 		return ret;
298 
299 	ret = ivpu_ipc_send_receive_internal(vdev, req, expected_resp_type, resp,
300 					     channel, timeout_ms);
301 	if (ret != -ETIMEDOUT)
302 		goto rpm_put;
303 
304 	hb_ret = ivpu_ipc_send_receive_internal(vdev, &hb_req, VPU_JSM_MSG_QUERY_ENGINE_HB_DONE,
305 						&hb_resp, VPU_IPC_CHAN_ASYNC_CMD,
306 						vdev->timeout.jsm);
307 	if (hb_ret == -ETIMEDOUT) {
308 		ivpu_hw_diagnose_failure(vdev);
309 		ivpu_pm_schedule_recovery(vdev);
310 	}
311 
312 rpm_put:
313 	ivpu_rpm_put(vdev);
314 	return ret;
315 }
316 
317 static bool
318 ivpu_ipc_match_consumer(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons,
319 			struct ivpu_ipc_hdr *ipc_hdr, struct vpu_jsm_msg *jsm_msg)
320 {
321 	if (cons->channel != ipc_hdr->channel)
322 		return false;
323 
324 	if (!jsm_msg || jsm_msg->request_id == cons->request_id)
325 		return true;
326 
327 	return false;
328 }
329 
330 static void
331 ivpu_ipc_dispatch(struct ivpu_device *vdev, struct ivpu_ipc_consumer *cons,
332 		  struct ivpu_ipc_hdr *ipc_hdr, struct vpu_jsm_msg *jsm_msg)
333 {
334 	struct ivpu_ipc_info *ipc = vdev->ipc;
335 	struct ivpu_ipc_rx_msg *rx_msg;
336 	unsigned long flags;
337 
338 	lockdep_assert_held(&ipc->cons_list_lock);
339 
340 	rx_msg = kzalloc(sizeof(*rx_msg), GFP_ATOMIC);
341 	if (!rx_msg) {
342 		ivpu_ipc_rx_mark_free(vdev, ipc_hdr, jsm_msg);
343 		return;
344 	}
345 
346 	atomic_inc(&ipc->rx_msg_count);
347 
348 	rx_msg->ipc_hdr = ipc_hdr;
349 	rx_msg->jsm_msg = jsm_msg;
350 
351 	spin_lock_irqsave(&cons->rx_msg_lock, flags);
352 	list_add_tail(&rx_msg->link, &cons->rx_msg_list);
353 	spin_unlock_irqrestore(&cons->rx_msg_lock, flags);
354 
355 	wake_up(&cons->rx_msg_wq);
356 }
357 
358 int ivpu_ipc_irq_handler(struct ivpu_device *vdev)
359 {
360 	struct ivpu_ipc_info *ipc = vdev->ipc;
361 	struct ivpu_ipc_consumer *cons;
362 	struct ivpu_ipc_hdr *ipc_hdr;
363 	struct vpu_jsm_msg *jsm_msg;
364 	unsigned long flags;
365 	bool dispatched;
366 	u32 vpu_addr;
367 
368 	/*
369 	 * Driver needs to purge all messages from IPC FIFO to clear IPC interrupt.
370 	 * Without purge IPC FIFO to 0 next IPC interrupts won't be generated.
371 	 */
372 	while (ivpu_hw_reg_ipc_rx_count_get(vdev)) {
373 		vpu_addr = ivpu_hw_reg_ipc_rx_addr_get(vdev);
374 		if (vpu_addr == REG_IO_ERROR) {
375 			ivpu_err(vdev, "Failed to read IPC rx addr register\n");
376 			return -EIO;
377 		}
378 
379 		ipc_hdr = ivpu_to_cpu_addr(ipc->mem_rx, vpu_addr);
380 		if (!ipc_hdr) {
381 			ivpu_warn(vdev, "IPC msg 0x%x out of range\n", vpu_addr);
382 			continue;
383 		}
384 		ivpu_ipc_msg_dump(vdev, "RX", ipc_hdr, vpu_addr);
385 
386 		jsm_msg = NULL;
387 		if (ipc_hdr->channel != IVPU_IPC_CHAN_BOOT_MSG) {
388 			jsm_msg = ivpu_to_cpu_addr(ipc->mem_rx, ipc_hdr->data_addr);
389 			if (!jsm_msg) {
390 				ivpu_warn(vdev, "JSM msg 0x%x out of range\n", ipc_hdr->data_addr);
391 				ivpu_ipc_rx_mark_free(vdev, ipc_hdr, NULL);
392 				continue;
393 			}
394 			ivpu_jsm_msg_dump(vdev, "RX", jsm_msg, ipc_hdr->data_addr);
395 		}
396 
397 		if (atomic_read(&ipc->rx_msg_count) > IPC_MAX_RX_MSG) {
398 			ivpu_warn(vdev, "IPC RX msg dropped, msg count %d\n", IPC_MAX_RX_MSG);
399 			ivpu_ipc_rx_mark_free(vdev, ipc_hdr, jsm_msg);
400 			continue;
401 		}
402 
403 		dispatched = false;
404 		spin_lock_irqsave(&ipc->cons_list_lock, flags);
405 		list_for_each_entry(cons, &ipc->cons_list, link) {
406 			if (ivpu_ipc_match_consumer(vdev, cons, ipc_hdr, jsm_msg)) {
407 				ivpu_ipc_dispatch(vdev, cons, ipc_hdr, jsm_msg);
408 				dispatched = true;
409 				break;
410 			}
411 		}
412 		spin_unlock_irqrestore(&ipc->cons_list_lock, flags);
413 
414 		if (!dispatched) {
415 			ivpu_dbg(vdev, IPC, "IPC RX msg 0x%x dropped (no consumer)\n", vpu_addr);
416 			ivpu_ipc_rx_mark_free(vdev, ipc_hdr, jsm_msg);
417 		}
418 	}
419 
420 	return 0;
421 }
422 
423 int ivpu_ipc_init(struct ivpu_device *vdev)
424 {
425 	struct ivpu_ipc_info *ipc = vdev->ipc;
426 	int ret = -ENOMEM;
427 
428 	ipc->mem_tx = ivpu_bo_alloc_internal(vdev, 0, SZ_16K, DRM_IVPU_BO_WC);
429 	if (!ipc->mem_tx)
430 		return ret;
431 
432 	ipc->mem_rx = ivpu_bo_alloc_internal(vdev, 0, SZ_16K, DRM_IVPU_BO_WC);
433 	if (!ipc->mem_rx)
434 		goto err_free_tx;
435 
436 	ipc->mm_tx = devm_gen_pool_create(vdev->drm.dev, __ffs(IVPU_IPC_ALIGNMENT),
437 					  -1, "TX_IPC_JSM");
438 	if (IS_ERR(ipc->mm_tx)) {
439 		ret = PTR_ERR(ipc->mm_tx);
440 		ivpu_err(vdev, "Failed to create gen pool, %pe\n", ipc->mm_tx);
441 		goto err_free_rx;
442 	}
443 
444 	ret = gen_pool_add(ipc->mm_tx, ipc->mem_tx->vpu_addr, ipc->mem_tx->base.size, -1);
445 	if (ret) {
446 		ivpu_err(vdev, "gen_pool_add failed, ret %d\n", ret);
447 		goto err_free_rx;
448 	}
449 
450 	INIT_LIST_HEAD(&ipc->cons_list);
451 	spin_lock_init(&ipc->cons_list_lock);
452 	drmm_mutex_init(&vdev->drm, &ipc->lock);
453 
454 	ivpu_ipc_reset(vdev);
455 	return 0;
456 
457 err_free_rx:
458 	ivpu_bo_free_internal(ipc->mem_rx);
459 err_free_tx:
460 	ivpu_bo_free_internal(ipc->mem_tx);
461 	return ret;
462 }
463 
464 void ivpu_ipc_fini(struct ivpu_device *vdev)
465 {
466 	ivpu_ipc_mem_fini(vdev);
467 }
468 
469 void ivpu_ipc_enable(struct ivpu_device *vdev)
470 {
471 	struct ivpu_ipc_info *ipc = vdev->ipc;
472 
473 	mutex_lock(&ipc->lock);
474 	ipc->on = true;
475 	mutex_unlock(&ipc->lock);
476 }
477 
478 void ivpu_ipc_disable(struct ivpu_device *vdev)
479 {
480 	struct ivpu_ipc_info *ipc = vdev->ipc;
481 	struct ivpu_ipc_consumer *cons, *c;
482 	unsigned long flags;
483 
484 	mutex_lock(&ipc->lock);
485 	ipc->on = false;
486 	mutex_unlock(&ipc->lock);
487 
488 	spin_lock_irqsave(&ipc->cons_list_lock, flags);
489 	list_for_each_entry_safe(cons, c, &ipc->cons_list, link)
490 		wake_up(&cons->rx_msg_wq);
491 	spin_unlock_irqrestore(&ipc->cons_list_lock, flags);
492 }
493 
494 void ivpu_ipc_reset(struct ivpu_device *vdev)
495 {
496 	struct ivpu_ipc_info *ipc = vdev->ipc;
497 
498 	mutex_lock(&ipc->lock);
499 
500 	memset(ipc->mem_tx->kvaddr, 0, ipc->mem_tx->base.size);
501 	memset(ipc->mem_rx->kvaddr, 0, ipc->mem_rx->base.size);
502 	wmb(); /* Flush WC buffers for TX and RX rings */
503 
504 	mutex_unlock(&ipc->lock);
505 }
506