1 // SPDX-License-Identifier: GPL-2.0 OR MIT
2 /*
3  * Copyright 2016 VMware, Inc., Palo Alto, CA., USA
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the
7  * "Software"), to deal in the Software without restriction, including
8  * without limitation the rights to use, copy, modify, merge, publish,
9  * distribute, sub license, and/or sell copies of the Software, and to
10  * permit persons to whom the Software is furnished to do so, subject to
11  * the following conditions:
12  *
13  * The above copyright notice and this permission notice (including the
14  * next paragraph) shall be included in all copies or substantial portions
15  * of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
20  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
21  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
22  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
23  * USE OR OTHER DEALINGS IN THE SOFTWARE.
24  *
25  */
26 
27 #include <linux/objtool.h>
28 #include <linux/kernel.h>
29 #include <linux/module.h>
30 #include <linux/slab.h>
31 #include <linux/cc_platform.h>
32 
33 #include <asm/hypervisor.h>
34 #include <drm/drm_ioctl.h>
35 
36 #include "vmwgfx_drv.h"
37 #include "vmwgfx_msg_x86.h"
38 #include "vmwgfx_msg_arm64.h"
39 #include "vmwgfx_mksstat.h"
40 
41 #define MESSAGE_STATUS_SUCCESS  0x0001
42 #define MESSAGE_STATUS_DORECV   0x0002
43 #define MESSAGE_STATUS_CPT      0x0010
44 #define MESSAGE_STATUS_HB       0x0080
45 
46 #define RPCI_PROTOCOL_NUM       0x49435052
47 #define GUESTMSG_FLAG_COOKIE    0x80000000
48 
49 #define RETRIES                 3
50 
51 #define VMW_HYPERVISOR_MAGIC    0x564D5868
52 
53 #define VMW_PORT_CMD_MSG        30
54 #define VMW_PORT_CMD_HB_MSG     0
55 #define VMW_PORT_CMD_OPEN_CHANNEL  (MSG_TYPE_OPEN << 16 | VMW_PORT_CMD_MSG)
56 #define VMW_PORT_CMD_CLOSE_CHANNEL (MSG_TYPE_CLOSE << 16 | VMW_PORT_CMD_MSG)
57 #define VMW_PORT_CMD_SENDSIZE   (MSG_TYPE_SENDSIZE << 16 | VMW_PORT_CMD_MSG)
58 #define VMW_PORT_CMD_RECVSIZE   (MSG_TYPE_RECVSIZE << 16 | VMW_PORT_CMD_MSG)
59 #define VMW_PORT_CMD_RECVSTATUS (MSG_TYPE_RECVSTATUS << 16 | VMW_PORT_CMD_MSG)
60 
61 #define VMW_PORT_CMD_MKS_GUEST_STATS   85
62 #define VMW_PORT_CMD_MKSGS_RESET       (0 << 16 | VMW_PORT_CMD_MKS_GUEST_STATS)
63 #define VMW_PORT_CMD_MKSGS_ADD_PPN     (1 << 16 | VMW_PORT_CMD_MKS_GUEST_STATS)
64 #define VMW_PORT_CMD_MKSGS_REMOVE_PPN  (2 << 16 | VMW_PORT_CMD_MKS_GUEST_STATS)
65 
66 #define HIGH_WORD(X) ((X & 0xFFFF0000) >> 16)
67 
68 #define MAX_USER_MSG_LENGTH	PAGE_SIZE
69 
70 static u32 vmw_msg_enabled = 1;
71 
72 enum rpc_msg_type {
73 	MSG_TYPE_OPEN,
74 	MSG_TYPE_SENDSIZE,
75 	MSG_TYPE_SENDPAYLOAD,
76 	MSG_TYPE_RECVSIZE,
77 	MSG_TYPE_RECVPAYLOAD,
78 	MSG_TYPE_RECVSTATUS,
79 	MSG_TYPE_CLOSE,
80 };
81 
82 struct rpc_channel {
83 	u16 channel_id;
84 	u32 cookie_high;
85 	u32 cookie_low;
86 };
87 
88 #if IS_ENABLED(CONFIG_DRM_VMWGFX_MKSSTATS)
89 /* Kernel mksGuestStats counter names and desciptions; same order as enum mksstat_kern_stats_t */
90 static const char* const mksstat_kern_name_desc[MKSSTAT_KERN_COUNT][2] =
91 {
92 	{ "vmw_execbuf_ioctl", "vmw_execbuf_ioctl" },
93 	{ "vmw_cotable_resize", "vmw_cotable_resize" },
94 };
95 #endif
96 
97 /**
98  * vmw_open_channel
99  *
100  * @channel: RPC channel
101  * @protocol:
102  *
103  * Returns: 0 on success
104  */
105 static int vmw_open_channel(struct rpc_channel *channel, unsigned int protocol)
106 {
107 	unsigned long eax, ebx, ecx, edx, si = 0, di = 0;
108 
109 	VMW_PORT(VMW_PORT_CMD_OPEN_CHANNEL,
110 		(protocol | GUESTMSG_FLAG_COOKIE), si, di,
111 		0,
112 		VMW_HYPERVISOR_MAGIC,
113 		eax, ebx, ecx, edx, si, di);
114 
115 	if ((HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS) == 0)
116 		return -EINVAL;
117 
118 	channel->channel_id  = HIGH_WORD(edx);
119 	channel->cookie_high = si;
120 	channel->cookie_low  = di;
121 
122 	return 0;
123 }
124 
125 
126 
127 /**
128  * vmw_close_channel
129  *
130  * @channel: RPC channel
131  *
132  * Returns: 0 on success
133  */
134 static int vmw_close_channel(struct rpc_channel *channel)
135 {
136 	unsigned long eax, ebx, ecx, edx, si, di;
137 
138 	/* Set up additional parameters */
139 	si  = channel->cookie_high;
140 	di  = channel->cookie_low;
141 
142 	VMW_PORT(VMW_PORT_CMD_CLOSE_CHANNEL,
143 		0, si, di,
144 		channel->channel_id << 16,
145 		VMW_HYPERVISOR_MAGIC,
146 		eax, ebx, ecx, edx, si, di);
147 
148 	if ((HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS) == 0)
149 		return -EINVAL;
150 
151 	return 0;
152 }
153 
154 /**
155  * vmw_port_hb_out - Send the message payload either through the
156  * high-bandwidth port if available, or through the backdoor otherwise.
157  * @channel: The rpc channel.
158  * @msg: NULL-terminated message.
159  * @hb: Whether the high-bandwidth port is available.
160  *
161  * Return: The port status.
162  */
163 static unsigned long vmw_port_hb_out(struct rpc_channel *channel,
164 				     const char *msg, bool hb)
165 {
166 	unsigned long si, di, eax, ebx, ecx, edx;
167 	unsigned long msg_len = strlen(msg);
168 
169 	/* HB port can't access encrypted memory. */
170 	if (hb && !cc_platform_has(CC_ATTR_MEM_ENCRYPT)) {
171 		unsigned long bp = channel->cookie_high;
172 		u32 channel_id = (channel->channel_id << 16);
173 
174 		si = (uintptr_t) msg;
175 		di = channel->cookie_low;
176 
177 		VMW_PORT_HB_OUT(
178 			(MESSAGE_STATUS_SUCCESS << 16) | VMW_PORT_CMD_HB_MSG,
179 			msg_len, si, di,
180 			VMWARE_HYPERVISOR_HB | channel_id |
181 			VMWARE_HYPERVISOR_OUT,
182 			VMW_HYPERVISOR_MAGIC, bp,
183 			eax, ebx, ecx, edx, si, di);
184 
185 		return ebx;
186 	}
187 
188 	/* HB port not available. Send the message 4 bytes at a time. */
189 	ecx = MESSAGE_STATUS_SUCCESS << 16;
190 	while (msg_len && (HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS)) {
191 		unsigned int bytes = min_t(size_t, msg_len, 4);
192 		unsigned long word = 0;
193 
194 		memcpy(&word, msg, bytes);
195 		msg_len -= bytes;
196 		msg += bytes;
197 		si = channel->cookie_high;
198 		di = channel->cookie_low;
199 
200 		VMW_PORT(VMW_PORT_CMD_MSG | (MSG_TYPE_SENDPAYLOAD << 16),
201 			 word, si, di,
202 			 channel->channel_id << 16,
203 			 VMW_HYPERVISOR_MAGIC,
204 			 eax, ebx, ecx, edx, si, di);
205 	}
206 
207 	return ecx;
208 }
209 
210 /**
211  * vmw_port_hb_in - Receive the message payload either through the
212  * high-bandwidth port if available, or through the backdoor otherwise.
213  * @channel: The rpc channel.
214  * @reply: Pointer to buffer holding reply.
215  * @reply_len: Length of the reply.
216  * @hb: Whether the high-bandwidth port is available.
217  *
218  * Return: The port status.
219  */
220 static unsigned long vmw_port_hb_in(struct rpc_channel *channel, char *reply,
221 				    unsigned long reply_len, bool hb)
222 {
223 	unsigned long si, di, eax, ebx, ecx, edx;
224 
225 	/* HB port can't access encrypted memory */
226 	if (hb && !cc_platform_has(CC_ATTR_MEM_ENCRYPT)) {
227 		unsigned long bp = channel->cookie_low;
228 		u32 channel_id = (channel->channel_id << 16);
229 
230 		si = channel->cookie_high;
231 		di = (uintptr_t) reply;
232 
233 		VMW_PORT_HB_IN(
234 			(MESSAGE_STATUS_SUCCESS << 16) | VMW_PORT_CMD_HB_MSG,
235 			reply_len, si, di,
236 			VMWARE_HYPERVISOR_HB | channel_id,
237 			VMW_HYPERVISOR_MAGIC, bp,
238 			eax, ebx, ecx, edx, si, di);
239 
240 		return ebx;
241 	}
242 
243 	/* HB port not available. Retrieve the message 4 bytes at a time. */
244 	ecx = MESSAGE_STATUS_SUCCESS << 16;
245 	while (reply_len) {
246 		unsigned int bytes = min_t(unsigned long, reply_len, 4);
247 
248 		si = channel->cookie_high;
249 		di = channel->cookie_low;
250 
251 		VMW_PORT(VMW_PORT_CMD_MSG | (MSG_TYPE_RECVPAYLOAD << 16),
252 			 MESSAGE_STATUS_SUCCESS, si, di,
253 			 channel->channel_id << 16,
254 			 VMW_HYPERVISOR_MAGIC,
255 			 eax, ebx, ecx, edx, si, di);
256 
257 		if ((HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS) == 0)
258 			break;
259 
260 		memcpy(reply, &ebx, bytes);
261 		reply_len -= bytes;
262 		reply += bytes;
263 	}
264 
265 	return ecx;
266 }
267 
268 
269 /**
270  * vmw_send_msg: Sends a message to the host
271  *
272  * @channel: RPC channel
273  * @msg: NULL terminated string
274  *
275  * Returns: 0 on success
276  */
277 static int vmw_send_msg(struct rpc_channel *channel, const char *msg)
278 {
279 	unsigned long eax, ebx, ecx, edx, si, di;
280 	size_t msg_len = strlen(msg);
281 	int retries = 0;
282 
283 	while (retries < RETRIES) {
284 		retries++;
285 
286 		/* Set up additional parameters */
287 		si  = channel->cookie_high;
288 		di  = channel->cookie_low;
289 
290 		VMW_PORT(VMW_PORT_CMD_SENDSIZE,
291 			msg_len, si, di,
292 			channel->channel_id << 16,
293 			VMW_HYPERVISOR_MAGIC,
294 			eax, ebx, ecx, edx, si, di);
295 
296 		if ((HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS) == 0) {
297 			/* Expected success. Give up. */
298 			return -EINVAL;
299 		}
300 
301 		/* Send msg */
302 		ebx = vmw_port_hb_out(channel, msg,
303 				      !!(HIGH_WORD(ecx) & MESSAGE_STATUS_HB));
304 
305 		if ((HIGH_WORD(ebx) & MESSAGE_STATUS_SUCCESS) != 0) {
306 			return 0;
307 		} else if ((HIGH_WORD(ebx) & MESSAGE_STATUS_CPT) != 0) {
308 			/* A checkpoint occurred. Retry. */
309 			continue;
310 		} else {
311 			break;
312 		}
313 	}
314 
315 	return -EINVAL;
316 }
317 STACK_FRAME_NON_STANDARD(vmw_send_msg);
318 
319 
320 /**
321  * vmw_recv_msg: Receives a message from the host
322  *
323  * Note:  It is the caller's responsibility to call kfree() on msg.
324  *
325  * @channel:  channel opened by vmw_open_channel
326  * @msg:  [OUT] message received from the host
327  * @msg_len: message length
328  */
329 static int vmw_recv_msg(struct rpc_channel *channel, void **msg,
330 			size_t *msg_len)
331 {
332 	unsigned long eax, ebx, ecx, edx, si, di;
333 	char *reply;
334 	size_t reply_len;
335 	int retries = 0;
336 
337 
338 	*msg_len = 0;
339 	*msg = NULL;
340 
341 	while (retries < RETRIES) {
342 		retries++;
343 
344 		/* Set up additional parameters */
345 		si  = channel->cookie_high;
346 		di  = channel->cookie_low;
347 
348 		VMW_PORT(VMW_PORT_CMD_RECVSIZE,
349 			0, si, di,
350 			channel->channel_id << 16,
351 			VMW_HYPERVISOR_MAGIC,
352 			eax, ebx, ecx, edx, si, di);
353 
354 		if ((HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS) == 0) {
355 			DRM_ERROR("Failed to get reply size for host message.\n");
356 			return -EINVAL;
357 		}
358 
359 		/* No reply available.  This is okay. */
360 		if ((HIGH_WORD(ecx) & MESSAGE_STATUS_DORECV) == 0)
361 			return 0;
362 
363 		reply_len = ebx;
364 		reply     = kzalloc(reply_len + 1, GFP_KERNEL);
365 		if (!reply) {
366 			DRM_ERROR("Cannot allocate memory for host message reply.\n");
367 			return -ENOMEM;
368 		}
369 
370 
371 		/* Receive buffer */
372 		ebx = vmw_port_hb_in(channel, reply, reply_len,
373 				     !!(HIGH_WORD(ecx) & MESSAGE_STATUS_HB));
374 		if ((HIGH_WORD(ebx) & MESSAGE_STATUS_SUCCESS) == 0) {
375 			kfree(reply);
376 			reply = NULL;
377 			if ((HIGH_WORD(ebx) & MESSAGE_STATUS_CPT) != 0) {
378 				/* A checkpoint occurred. Retry. */
379 				continue;
380 			}
381 
382 			return -EINVAL;
383 		}
384 
385 		reply[reply_len] = '\0';
386 
387 
388 		/* Ack buffer */
389 		si  = channel->cookie_high;
390 		di  = channel->cookie_low;
391 
392 		VMW_PORT(VMW_PORT_CMD_RECVSTATUS,
393 			MESSAGE_STATUS_SUCCESS, si, di,
394 			channel->channel_id << 16,
395 			VMW_HYPERVISOR_MAGIC,
396 			eax, ebx, ecx, edx, si, di);
397 
398 		if ((HIGH_WORD(ecx) & MESSAGE_STATUS_SUCCESS) == 0) {
399 			kfree(reply);
400 			reply = NULL;
401 			if ((HIGH_WORD(ecx) & MESSAGE_STATUS_CPT) != 0) {
402 				/* A checkpoint occurred. Retry. */
403 				continue;
404 			}
405 
406 			return -EINVAL;
407 		}
408 
409 		break;
410 	}
411 
412 	if (!reply)
413 		return -EINVAL;
414 
415 	*msg_len = reply_len;
416 	*msg     = reply;
417 
418 	return 0;
419 }
420 STACK_FRAME_NON_STANDARD(vmw_recv_msg);
421 
422 
423 /**
424  * vmw_host_get_guestinfo: Gets a GuestInfo parameter
425  *
426  * Gets the value of a  GuestInfo.* parameter.  The value returned will be in
427  * a string, and it is up to the caller to post-process.
428  *
429  * @guest_info_param:  Parameter to get, e.g. GuestInfo.svga.gl3
430  * @buffer: if NULL, *reply_len will contain reply size.
431  * @length: size of the reply_buf.  Set to size of reply upon return
432  *
433  * Returns: 0 on success
434  */
435 int vmw_host_get_guestinfo(const char *guest_info_param,
436 			   char *buffer, size_t *length)
437 {
438 	struct rpc_channel channel;
439 	char *msg, *reply = NULL;
440 	size_t reply_len = 0;
441 
442 	if (!vmw_msg_enabled)
443 		return -ENODEV;
444 
445 	if (!guest_info_param || !length)
446 		return -EINVAL;
447 
448 	msg = kasprintf(GFP_KERNEL, "info-get %s", guest_info_param);
449 	if (!msg) {
450 		DRM_ERROR("Cannot allocate memory to get guest info \"%s\".",
451 			  guest_info_param);
452 		return -ENOMEM;
453 	}
454 
455 	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM))
456 		goto out_open;
457 
458 	if (vmw_send_msg(&channel, msg) ||
459 	    vmw_recv_msg(&channel, (void *) &reply, &reply_len))
460 		goto out_msg;
461 
462 	vmw_close_channel(&channel);
463 	if (buffer && reply && reply_len > 0) {
464 		/* Remove reply code, which are the first 2 characters of
465 		 * the reply
466 		 */
467 		reply_len = max(reply_len - 2, (size_t) 0);
468 		reply_len = min(reply_len, *length);
469 
470 		if (reply_len > 0)
471 			memcpy(buffer, reply + 2, reply_len);
472 	}
473 
474 	*length = reply_len;
475 
476 	kfree(reply);
477 	kfree(msg);
478 
479 	return 0;
480 
481 out_msg:
482 	vmw_close_channel(&channel);
483 	kfree(reply);
484 out_open:
485 	*length = 0;
486 	kfree(msg);
487 	DRM_ERROR("Failed to get guest info \"%s\".", guest_info_param);
488 
489 	return -EINVAL;
490 }
491 
492 
493 /**
494  * vmw_host_printf: Sends a log message to the host
495  *
496  * @fmt: Regular printf format string and arguments
497  *
498  * Returns: 0 on success
499  */
500 __printf(1, 2)
501 int vmw_host_printf(const char *fmt, ...)
502 {
503 	va_list ap;
504 	struct rpc_channel channel;
505 	char *msg;
506 	char *log;
507 	int ret = 0;
508 
509 	if (!vmw_msg_enabled)
510 		return -ENODEV;
511 
512 	if (!fmt)
513 		return ret;
514 
515 	va_start(ap, fmt);
516 	log = kvasprintf(GFP_KERNEL, fmt, ap);
517 	va_end(ap);
518 	if (!log) {
519 		DRM_ERROR("Cannot allocate memory for the log message.\n");
520 		return -ENOMEM;
521 	}
522 
523 	msg = kasprintf(GFP_KERNEL, "log %s", log);
524 	if (!msg) {
525 		DRM_ERROR("Cannot allocate memory for host log message.\n");
526 		kfree(log);
527 		return -ENOMEM;
528 	}
529 
530 	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM))
531 		goto out_open;
532 
533 	if (vmw_send_msg(&channel, msg))
534 		goto out_msg;
535 
536 	vmw_close_channel(&channel);
537 	kfree(msg);
538 	kfree(log);
539 
540 	return 0;
541 
542 out_msg:
543 	vmw_close_channel(&channel);
544 out_open:
545 	kfree(msg);
546 	kfree(log);
547 	DRM_ERROR("Failed to send host log message.\n");
548 
549 	return -EINVAL;
550 }
551 
552 
553 /**
554  * vmw_msg_ioctl: Sends and receveives a message to/from host from/to user-space
555  *
556  * Sends a message from user-space to host.
557  * Can also receive a result from host and return that to user-space.
558  *
559  * @dev: Identifies the drm device.
560  * @data: Pointer to the ioctl argument.
561  * @file_priv: Identifies the caller.
562  * Return: Zero on success, negative error code on error.
563  */
564 
565 int vmw_msg_ioctl(struct drm_device *dev, void *data,
566 		  struct drm_file *file_priv)
567 {
568 	struct drm_vmw_msg_arg *arg =
569 			(struct drm_vmw_msg_arg *)data;
570 	struct rpc_channel channel;
571 	char *msg;
572 	int length;
573 
574 	msg = kmalloc(MAX_USER_MSG_LENGTH, GFP_KERNEL);
575 	if (!msg) {
576 		DRM_ERROR("Cannot allocate memory for log message.\n");
577 		return -ENOMEM;
578 	}
579 
580 	length = strncpy_from_user(msg, (void __user *)((unsigned long)arg->send),
581 				   MAX_USER_MSG_LENGTH);
582 	if (length < 0 || length >= MAX_USER_MSG_LENGTH) {
583 		DRM_ERROR("Userspace message access failure.\n");
584 		kfree(msg);
585 		return -EINVAL;
586 	}
587 
588 
589 	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM)) {
590 		DRM_ERROR("Failed to open channel.\n");
591 		goto out_open;
592 	}
593 
594 	if (vmw_send_msg(&channel, msg)) {
595 		DRM_ERROR("Failed to send message to host.\n");
596 		goto out_msg;
597 	}
598 
599 	if (!arg->send_only) {
600 		char *reply = NULL;
601 		size_t reply_len = 0;
602 
603 		if (vmw_recv_msg(&channel, (void *) &reply, &reply_len)) {
604 			DRM_ERROR("Failed to receive message from host.\n");
605 			goto out_msg;
606 		}
607 		if (reply && reply_len > 0) {
608 			if (copy_to_user((void __user *)((unsigned long)arg->receive),
609 					 reply, reply_len)) {
610 				DRM_ERROR("Failed to copy message to userspace.\n");
611 				kfree(reply);
612 				goto out_msg;
613 			}
614 			arg->receive_len = (__u32)reply_len;
615 		}
616 		kfree(reply);
617 	}
618 
619 	vmw_close_channel(&channel);
620 	kfree(msg);
621 
622 	return 0;
623 
624 out_msg:
625 	vmw_close_channel(&channel);
626 out_open:
627 	kfree(msg);
628 
629 	return -EINVAL;
630 }
631 
632 /**
633  * reset_ppn_array: Resets a PPN64 array to INVALID_PPN64 content
634  *
635  * @arr: Array to reset.
636  * @size: Array length.
637  */
638 static inline void reset_ppn_array(PPN64 *arr, size_t size)
639 {
640 	size_t i;
641 
642 	BUG_ON(!arr || size == 0);
643 
644 	for (i = 0; i < size; ++i)
645 		arr[i] = INVALID_PPN64;
646 }
647 
648 /**
649  * hypervisor_ppn_reset_all: Removes all mksGuestStat instance descriptors from
650  * the hypervisor. All related pages should be subsequently unpinned or freed.
651  *
652  */
653 static inline void hypervisor_ppn_reset_all(void)
654 {
655 	unsigned long eax, ebx, ecx, edx, si = 0, di = 0;
656 
657 	VMW_PORT(VMW_PORT_CMD_MKSGS_RESET,
658 		0, si, di,
659 		0,
660 		VMW_HYPERVISOR_MAGIC,
661 		eax, ebx, ecx, edx, si, di);
662 }
663 
664 /**
665  * hypervisor_ppn_add: Adds a single mksGuestStat instance descriptor to the
666  * hypervisor. Any related userspace pages should be pinned in advance.
667  *
668  * @pfn: Physical page number of the instance descriptor
669  */
670 static inline void hypervisor_ppn_add(PPN64 pfn)
671 {
672 	unsigned long eax, ebx, ecx, edx, si = 0, di = 0;
673 
674 	VMW_PORT(VMW_PORT_CMD_MKSGS_ADD_PPN,
675 		(unsigned long)pfn, si, di,
676 		0,
677 		VMW_HYPERVISOR_MAGIC,
678 		eax, ebx, ecx, edx, si, di);
679 }
680 
681 /**
682  * hypervisor_ppn_remove: Removes a single mksGuestStat instance descriptor from
683  * the hypervisor. All related pages should be subsequently unpinned or freed.
684  *
685  * @pfn: Physical page number of the instance descriptor
686  */
687 static inline void hypervisor_ppn_remove(PPN64 pfn)
688 {
689 	unsigned long eax, ebx, ecx, edx, si = 0, di = 0;
690 
691 	VMW_PORT(VMW_PORT_CMD_MKSGS_REMOVE_PPN,
692 		(unsigned long)pfn, si, di,
693 		0,
694 		VMW_HYPERVISOR_MAGIC,
695 		eax, ebx, ecx, edx, si, di);
696 }
697 
698 #if IS_ENABLED(CONFIG_DRM_VMWGFX_MKSSTATS)
699 
700 /* Order of the total number of pages used for kernel-internal mksGuestStat; at least 2 */
701 #define MKSSTAT_KERNEL_PAGES_ORDER 2
702 /* Header to the text description of mksGuestStat instance descriptor */
703 #define MKSSTAT_KERNEL_DESCRIPTION "vmwgfx"
704 
705 /**
706  * mksstat_init_record_time: Initializes an MKSGuestStatCounterTime-based record
707  * for the respective mksGuestStat index.
708  *
709  * @stat_idx: Index of the MKSGuestStatCounterTime-based mksGuestStat record.
710  * @pstat: Pointer to array of MKSGuestStatCounterTime.
711  * @pinfo: Pointer to array of MKSGuestStatInfoEntry.
712  * @pstrs: Pointer to current end of the name/description sequence.
713  * Return: Pointer to the new end of the names/description sequence.
714  */
715 
716 static inline char *mksstat_init_record_time(mksstat_kern_stats_t stat_idx,
717 	MKSGuestStatCounterTime *pstat, MKSGuestStatInfoEntry *pinfo, char *pstrs)
718 {
719 	char *const pstrd = pstrs + strlen(mksstat_kern_name_desc[stat_idx][0]) + 1;
720 	strcpy(pstrs, mksstat_kern_name_desc[stat_idx][0]);
721 	strcpy(pstrd, mksstat_kern_name_desc[stat_idx][1]);
722 
723 	pinfo[stat_idx].name.s = pstrs;
724 	pinfo[stat_idx].description.s = pstrd;
725 	pinfo[stat_idx].flags = MKS_GUEST_STAT_FLAG_TIME;
726 	pinfo[stat_idx].stat.counterTime = &pstat[stat_idx];
727 
728 	return pstrd + strlen(mksstat_kern_name_desc[stat_idx][1]) + 1;
729 }
730 
731 /**
732  * mksstat_init_kern_id: Creates a single mksGuestStat instance descriptor and
733  * kernel-internal counters. Adds PFN mapping to the hypervisor.
734  *
735  * Create a single mksGuestStat instance descriptor and corresponding structures
736  * for all kernel-internal counters. The corresponding PFNs are mapped with the
737  * hypervisor.
738  *
739  * @ppage: Output pointer to page containing the instance descriptor.
740  * Return: Zero on success, negative error code on error.
741  */
742 
743 static int mksstat_init_kern_id(struct page **ppage)
744 {
745 	MKSGuestStatInstanceDescriptor *pdesc;
746 	MKSGuestStatCounterTime *pstat;
747 	MKSGuestStatInfoEntry *pinfo;
748 	char *pstrs, *pstrs_acc;
749 
750 	/* Allocate pages for the kernel-internal instance descriptor */
751 	struct page *page = alloc_pages(GFP_KERNEL | __GFP_ZERO, MKSSTAT_KERNEL_PAGES_ORDER);
752 
753 	if (!page)
754 		return -ENOMEM;
755 
756 	pdesc = page_address(page);
757 	pstat = vmw_mksstat_get_kern_pstat(pdesc);
758 	pinfo = vmw_mksstat_get_kern_pinfo(pdesc);
759 	pstrs = vmw_mksstat_get_kern_pstrs(pdesc);
760 
761 	/* Set up all kernel-internal counters and corresponding structures */
762 	pstrs_acc = pstrs;
763 	pstrs_acc = mksstat_init_record_time(MKSSTAT_KERN_EXECBUF, pstat, pinfo, pstrs_acc);
764 	pstrs_acc = mksstat_init_record_time(MKSSTAT_KERN_COTABLE_RESIZE, pstat, pinfo, pstrs_acc);
765 
766 	/* Add new counters above, in their order of appearance in mksstat_kern_stats_t */
767 
768 	BUG_ON(pstrs_acc - pstrs > PAGE_SIZE);
769 
770 	/* Set up the kernel-internal instance descriptor */
771 	pdesc->reservedMBZ = 0;
772 	pdesc->statStartVA = (uintptr_t)pstat;
773 	pdesc->strsStartVA = (uintptr_t)pstrs;
774 	pdesc->statLength = sizeof(*pstat) * MKSSTAT_KERN_COUNT;
775 	pdesc->infoLength = sizeof(*pinfo) * MKSSTAT_KERN_COUNT;
776 	pdesc->strsLength = pstrs_acc - pstrs;
777 	snprintf(pdesc->description, ARRAY_SIZE(pdesc->description) - 1, "%s pid=%d",
778 		MKSSTAT_KERNEL_DESCRIPTION, current->pid);
779 
780 	pdesc->statPPNs[0] = page_to_pfn(virt_to_page(pstat));
781 	reset_ppn_array(pdesc->statPPNs + 1, ARRAY_SIZE(pdesc->statPPNs) - 1);
782 
783 	pdesc->infoPPNs[0] = page_to_pfn(virt_to_page(pinfo));
784 	reset_ppn_array(pdesc->infoPPNs + 1, ARRAY_SIZE(pdesc->infoPPNs) - 1);
785 
786 	pdesc->strsPPNs[0] = page_to_pfn(virt_to_page(pstrs));
787 	reset_ppn_array(pdesc->strsPPNs + 1, ARRAY_SIZE(pdesc->strsPPNs) - 1);
788 
789 	*ppage = page;
790 
791 	hypervisor_ppn_add((PPN64)page_to_pfn(page));
792 
793 	return 0;
794 }
795 
796 /**
797  * vmw_mksstat_get_kern_slot: Acquires a slot for a single kernel-internal
798  * mksGuestStat instance descriptor.
799  *
800  * Find a slot for a single kernel-internal mksGuestStat instance descriptor.
801  * In case no such was already present, allocate a new one and set up a kernel-
802  * internal mksGuestStat instance descriptor for the former.
803  *
804  * @pid: Process for which a slot is sought.
805  * @dev_priv: Identifies the drm private device.
806  * Return: Non-negative slot on success, negative error code on error.
807  */
808 
809 int vmw_mksstat_get_kern_slot(pid_t pid, struct vmw_private *dev_priv)
810 {
811 	const size_t base = (u32)hash_32(pid, MKSSTAT_CAPACITY_LOG2);
812 	size_t i;
813 
814 	for (i = 0; i < ARRAY_SIZE(dev_priv->mksstat_kern_pids); ++i) {
815 		const size_t slot = (i + base) % ARRAY_SIZE(dev_priv->mksstat_kern_pids);
816 
817 		/* Check if an instance descriptor for this pid is already present */
818 		if (pid == (pid_t)atomic_read(&dev_priv->mksstat_kern_pids[slot]))
819 			return (int)slot;
820 
821 		/* Set up a new instance descriptor for this pid */
822 		if (!atomic_cmpxchg(&dev_priv->mksstat_kern_pids[slot], 0, MKSSTAT_PID_RESERVED)) {
823 			const int ret = mksstat_init_kern_id(&dev_priv->mksstat_kern_pages[slot]);
824 
825 			if (!ret) {
826 				/* Reset top-timer tracking for this slot */
827 				dev_priv->mksstat_kern_top_timer[slot] = MKSSTAT_KERN_COUNT;
828 
829 				atomic_set(&dev_priv->mksstat_kern_pids[slot], pid);
830 				return (int)slot;
831 			}
832 
833 			atomic_set(&dev_priv->mksstat_kern_pids[slot], 0);
834 			return ret;
835 		}
836 	}
837 
838 	return -ENOSPC;
839 }
840 
841 #endif
842 
843 /**
844  * vmw_mksstat_cleanup_descriptor: Frees a single userspace-originating
845  * mksGuestStat instance-descriptor page and unpins all related user pages.
846  *
847  * Unpin all user pages realated to this instance descriptor and free
848  * the instance-descriptor page itself.
849  *
850  * @page: Page of the instance descriptor.
851  */
852 
853 static void vmw_mksstat_cleanup_descriptor(struct page *page)
854 {
855 	MKSGuestStatInstanceDescriptor *pdesc = page_address(page);
856 	size_t i;
857 
858 	for (i = 0; i < ARRAY_SIZE(pdesc->statPPNs) && pdesc->statPPNs[i] != INVALID_PPN64; ++i)
859 		unpin_user_page(pfn_to_page(pdesc->statPPNs[i]));
860 
861 	for (i = 0; i < ARRAY_SIZE(pdesc->infoPPNs) && pdesc->infoPPNs[i] != INVALID_PPN64; ++i)
862 		unpin_user_page(pfn_to_page(pdesc->infoPPNs[i]));
863 
864 	for (i = 0; i < ARRAY_SIZE(pdesc->strsPPNs) && pdesc->strsPPNs[i] != INVALID_PPN64; ++i)
865 		unpin_user_page(pfn_to_page(pdesc->strsPPNs[i]));
866 
867 	__free_page(page);
868 }
869 
870 /**
871  * vmw_mksstat_remove_all: Resets all mksGuestStat instance descriptors
872  * from the hypervisor.
873  *
874  * Discard all hypervisor PFN mappings, containing active mksGuestState instance
875  * descriptors, unpin the related userspace pages and free the related kernel pages.
876  *
877  * @dev_priv: Identifies the drm private device.
878  * Return: Zero on success, negative error code on error.
879  */
880 
881 int vmw_mksstat_remove_all(struct vmw_private *dev_priv)
882 {
883 	int ret = 0;
884 	size_t i;
885 
886 	/* Discard all PFN mappings with the hypervisor */
887 	hypervisor_ppn_reset_all();
888 
889 	/* Discard all userspace-originating instance descriptors and unpin all related pages */
890 	for (i = 0; i < ARRAY_SIZE(dev_priv->mksstat_user_pids); ++i) {
891 		const pid_t pid0 = (pid_t)atomic_read(&dev_priv->mksstat_user_pids[i]);
892 
893 		if (!pid0)
894 			continue;
895 
896 		if (pid0 != MKSSTAT_PID_RESERVED) {
897 			const pid_t pid1 = atomic_cmpxchg(&dev_priv->mksstat_user_pids[i], pid0, MKSSTAT_PID_RESERVED);
898 
899 			if (!pid1)
900 				continue;
901 
902 			if (pid1 == pid0) {
903 				struct page *const page = dev_priv->mksstat_user_pages[i];
904 
905 				BUG_ON(!page);
906 
907 				dev_priv->mksstat_user_pages[i] = NULL;
908 				atomic_set(&dev_priv->mksstat_user_pids[i], 0);
909 
910 				vmw_mksstat_cleanup_descriptor(page);
911 				continue;
912 			}
913 		}
914 
915 		ret = -EAGAIN;
916 	}
917 
918 #if IS_ENABLED(CONFIG_DRM_VMWGFX_MKSSTATS)
919 	/* Discard all kernel-internal instance descriptors and free all related pages */
920 	for (i = 0; i < ARRAY_SIZE(dev_priv->mksstat_kern_pids); ++i) {
921 		const pid_t pid0 = (pid_t)atomic_read(&dev_priv->mksstat_kern_pids[i]);
922 
923 		if (!pid0)
924 			continue;
925 
926 		if (pid0 != MKSSTAT_PID_RESERVED) {
927 			const pid_t pid1 = atomic_cmpxchg(&dev_priv->mksstat_kern_pids[i], pid0, MKSSTAT_PID_RESERVED);
928 
929 			if (!pid1)
930 				continue;
931 
932 			if (pid1 == pid0) {
933 				struct page *const page = dev_priv->mksstat_kern_pages[i];
934 
935 				BUG_ON(!page);
936 
937 				dev_priv->mksstat_kern_pages[i] = NULL;
938 				atomic_set(&dev_priv->mksstat_kern_pids[i], 0);
939 
940 				__free_pages(page, MKSSTAT_KERNEL_PAGES_ORDER);
941 				continue;
942 			}
943 		}
944 
945 		ret = -EAGAIN;
946 	}
947 
948 #endif
949 	return ret;
950 }
951 
952 /**
953  * vmw_mksstat_reset_ioctl: Resets all mksGuestStat instance descriptors
954  * from the hypervisor.
955  *
956  * Discard all hypervisor PFN mappings, containing active mksGuestStat instance
957  * descriptors, unpin the related userspace pages and free the related kernel pages.
958  *
959  * @dev: Identifies the drm device.
960  * @data: Pointer to the ioctl argument.
961  * @file_priv: Identifies the caller; unused.
962  * Return: Zero on success, negative error code on error.
963  */
964 
965 int vmw_mksstat_reset_ioctl(struct drm_device *dev, void *data,
966 				struct drm_file *file_priv)
967 {
968 	struct vmw_private *const dev_priv = vmw_priv(dev);
969 	return vmw_mksstat_remove_all(dev_priv);
970 }
971 
972 /**
973  * vmw_mksstat_add_ioctl: Creates a single userspace-originating mksGuestStat
974  * instance descriptor and registers that with the hypervisor.
975  *
976  * Create a hypervisor PFN mapping, containing a single mksGuestStat instance
977  * descriptor and pin the corresponding userspace pages.
978  *
979  * @dev: Identifies the drm device.
980  * @data: Pointer to the ioctl argument.
981  * @file_priv: Identifies the caller; unused.
982  * Return: Zero on success, negative error code on error.
983  */
984 
985 int vmw_mksstat_add_ioctl(struct drm_device *dev, void *data,
986 				struct drm_file *file_priv)
987 {
988 	struct drm_vmw_mksstat_add_arg *arg =
989 		(struct drm_vmw_mksstat_add_arg *) data;
990 
991 	struct vmw_private *const dev_priv = vmw_priv(dev);
992 
993 	const size_t num_pages_stat = PFN_UP(arg->stat_len);
994 	const size_t num_pages_info = PFN_UP(arg->info_len);
995 	const size_t num_pages_strs = PFN_UP(arg->strs_len);
996 	long desc_len;
997 	long nr_pinned_stat;
998 	long nr_pinned_info;
999 	long nr_pinned_strs;
1000 	MKSGuestStatInstanceDescriptor *pdesc;
1001 	struct page *page = NULL;
1002 	struct page **pages_stat = NULL;
1003 	struct page **pages_info = NULL;
1004 	struct page **pages_strs = NULL;
1005 	size_t i, slot;
1006 	int ret_err = -ENOMEM;
1007 
1008 	arg->id = -1;
1009 
1010 	if (!arg->stat || !arg->info || !arg->strs)
1011 		return -EINVAL;
1012 
1013 	if (!arg->stat_len || !arg->info_len || !arg->strs_len)
1014 		return -EINVAL;
1015 
1016 	if (!arg->description)
1017 		return -EINVAL;
1018 
1019 	if (num_pages_stat > ARRAY_SIZE(pdesc->statPPNs) ||
1020 		num_pages_info > ARRAY_SIZE(pdesc->infoPPNs) ||
1021 		num_pages_strs > ARRAY_SIZE(pdesc->strsPPNs))
1022 		return -EINVAL;
1023 
1024 	/* Find an available slot in the mksGuestStats user array and reserve it */
1025 	for (slot = 0; slot < ARRAY_SIZE(dev_priv->mksstat_user_pids); ++slot)
1026 		if (!atomic_cmpxchg(&dev_priv->mksstat_user_pids[slot], 0, MKSSTAT_PID_RESERVED))
1027 			break;
1028 
1029 	if (slot == ARRAY_SIZE(dev_priv->mksstat_user_pids))
1030 		return -ENOSPC;
1031 
1032 	BUG_ON(dev_priv->mksstat_user_pages[slot]);
1033 
1034 	/* Allocate statically-sized temp arrays for pages -- too big to keep in frame */
1035 	pages_stat = (struct page **)kmalloc_array(
1036 		ARRAY_SIZE(pdesc->statPPNs) +
1037 		ARRAY_SIZE(pdesc->infoPPNs) +
1038 		ARRAY_SIZE(pdesc->strsPPNs), sizeof(*pages_stat), GFP_KERNEL);
1039 
1040 	if (!pages_stat)
1041 		goto err_nomem;
1042 
1043 	pages_info = pages_stat + ARRAY_SIZE(pdesc->statPPNs);
1044 	pages_strs = pages_info + ARRAY_SIZE(pdesc->infoPPNs);
1045 
1046 	/* Allocate a page for the instance descriptor */
1047 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1048 
1049 	if (!page)
1050 		goto err_nomem;
1051 
1052 	/* Set up the instance descriptor */
1053 	pdesc = page_address(page);
1054 
1055 	pdesc->reservedMBZ = 0;
1056 	pdesc->statStartVA = arg->stat;
1057 	pdesc->strsStartVA = arg->strs;
1058 	pdesc->statLength = arg->stat_len;
1059 	pdesc->infoLength = arg->info_len;
1060 	pdesc->strsLength = arg->strs_len;
1061 	desc_len = strncpy_from_user(pdesc->description, u64_to_user_ptr(arg->description),
1062 		ARRAY_SIZE(pdesc->description) - 1);
1063 
1064 	if (desc_len < 0) {
1065 		ret_err = -EFAULT;
1066 		goto err_nomem;
1067 	}
1068 
1069 	reset_ppn_array(pdesc->statPPNs, ARRAY_SIZE(pdesc->statPPNs));
1070 	reset_ppn_array(pdesc->infoPPNs, ARRAY_SIZE(pdesc->infoPPNs));
1071 	reset_ppn_array(pdesc->strsPPNs, ARRAY_SIZE(pdesc->strsPPNs));
1072 
1073 	/* Pin mksGuestStat user pages and store those in the instance descriptor */
1074 	nr_pinned_stat = pin_user_pages_fast(arg->stat, num_pages_stat, FOLL_LONGTERM, pages_stat);
1075 	if (num_pages_stat != nr_pinned_stat)
1076 		goto err_pin_stat;
1077 
1078 	for (i = 0; i < num_pages_stat; ++i)
1079 		pdesc->statPPNs[i] = page_to_pfn(pages_stat[i]);
1080 
1081 	nr_pinned_info = pin_user_pages_fast(arg->info, num_pages_info, FOLL_LONGTERM, pages_info);
1082 	if (num_pages_info != nr_pinned_info)
1083 		goto err_pin_info;
1084 
1085 	for (i = 0; i < num_pages_info; ++i)
1086 		pdesc->infoPPNs[i] = page_to_pfn(pages_info[i]);
1087 
1088 	nr_pinned_strs = pin_user_pages_fast(arg->strs, num_pages_strs, FOLL_LONGTERM, pages_strs);
1089 	if (num_pages_strs != nr_pinned_strs)
1090 		goto err_pin_strs;
1091 
1092 	for (i = 0; i < num_pages_strs; ++i)
1093 		pdesc->strsPPNs[i] = page_to_pfn(pages_strs[i]);
1094 
1095 	/* Send the descriptor to the host via a hypervisor call. The mksGuestStat
1096 	   pages will remain in use until the user requests a matching remove stats
1097 	   or a stats reset occurs. */
1098 	hypervisor_ppn_add((PPN64)page_to_pfn(page));
1099 
1100 	dev_priv->mksstat_user_pages[slot] = page;
1101 	atomic_set(&dev_priv->mksstat_user_pids[slot], task_pgrp_vnr(current));
1102 
1103 	arg->id = slot;
1104 
1105 	DRM_DEV_INFO(dev->dev, "pid=%d arg.description='%.*s' id=%zu\n", current->pid, (int)desc_len, pdesc->description, slot);
1106 
1107 	kfree(pages_stat);
1108 	return 0;
1109 
1110 err_pin_strs:
1111 	if (nr_pinned_strs > 0)
1112 		unpin_user_pages(pages_strs, nr_pinned_strs);
1113 
1114 err_pin_info:
1115 	if (nr_pinned_info > 0)
1116 		unpin_user_pages(pages_info, nr_pinned_info);
1117 
1118 err_pin_stat:
1119 	if (nr_pinned_stat > 0)
1120 		unpin_user_pages(pages_stat, nr_pinned_stat);
1121 
1122 err_nomem:
1123 	atomic_set(&dev_priv->mksstat_user_pids[slot], 0);
1124 	if (page)
1125 		__free_page(page);
1126 	kfree(pages_stat);
1127 
1128 	return ret_err;
1129 }
1130 
1131 /**
1132  * vmw_mksstat_remove_ioctl: Removes a single userspace-originating mksGuestStat
1133  * instance descriptor from the hypervisor.
1134  *
1135  * Discard a hypervisor PFN mapping, containing a single mksGuestStat instance
1136  * descriptor and unpin the corresponding userspace pages.
1137  *
1138  * @dev: Identifies the drm device.
1139  * @data: Pointer to the ioctl argument.
1140  * @file_priv: Identifies the caller; unused.
1141  * Return: Zero on success, negative error code on error.
1142  */
1143 
1144 int vmw_mksstat_remove_ioctl(struct drm_device *dev, void *data,
1145 				struct drm_file *file_priv)
1146 {
1147 	struct drm_vmw_mksstat_remove_arg *arg =
1148 		(struct drm_vmw_mksstat_remove_arg *) data;
1149 
1150 	struct vmw_private *const dev_priv = vmw_priv(dev);
1151 
1152 	const size_t slot = arg->id;
1153 	pid_t pgid, pid;
1154 
1155 	if (slot >= ARRAY_SIZE(dev_priv->mksstat_user_pids))
1156 		return -EINVAL;
1157 
1158 	DRM_DEV_INFO(dev->dev, "pid=%d arg.id=%zu\n", current->pid, slot);
1159 
1160 	pgid = task_pgrp_vnr(current);
1161 	pid = atomic_cmpxchg(&dev_priv->mksstat_user_pids[slot], pgid, MKSSTAT_PID_RESERVED);
1162 
1163 	if (!pid)
1164 		return 0;
1165 
1166 	if (pid == pgid) {
1167 		struct page *const page = dev_priv->mksstat_user_pages[slot];
1168 
1169 		BUG_ON(!page);
1170 
1171 		dev_priv->mksstat_user_pages[slot] = NULL;
1172 		atomic_set(&dev_priv->mksstat_user_pids[slot], 0);
1173 
1174 		hypervisor_ppn_remove((PPN64)page_to_pfn(page));
1175 
1176 		vmw_mksstat_cleanup_descriptor(page);
1177 		return 0;
1178 	}
1179 
1180 	return -EAGAIN;
1181 }
1182 
1183 /**
1184  * vmw_disable_backdoor: Disables all backdoor communication
1185  * with the hypervisor.
1186  */
1187 void vmw_disable_backdoor(void)
1188 {
1189 	vmw_msg_enabled = 0;
1190 }
1191