xref: /linux/drivers/cxl/core/mbox.c (revision 8457669db968c98edb781892d73fa559e1efcbd4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright(c) 2020 Intel Corporation. All rights reserved. */
3 #include <linux/security.h>
4 #include <linux/debugfs.h>
5 #include <linux/ktime.h>
6 #include <linux/mutex.h>
7 #include <linux/unaligned.h>
8 #include <cxlpci.h>
9 #include <cxlmem.h>
10 #include <cxl.h>
11 
12 #include "core.h"
13 #include "trace.h"
14 #include "mce.h"
15 
16 static bool cxl_raw_allow_all;
17 
18 /**
19  * DOC: cxl mbox
20  *
21  * Core implementation of the CXL 2.0 Type-3 Memory Device Mailbox. The
22  * implementation is used by the cxl_pci driver to initialize the device
23  * and implement the cxl_mem.h IOCTL UAPI. It also implements the
24  * backend of the cxl_pmem_ctl() transport for LIBNVDIMM.
25  */
26 
27 #define cxl_for_each_cmd(cmd)                                                  \
28 	for ((cmd) = &cxl_mem_commands[0];                                     \
29 	     ((cmd) - cxl_mem_commands) < ARRAY_SIZE(cxl_mem_commands); (cmd)++)
30 
31 #define CXL_CMD(_id, sin, sout, _flags)                                        \
32 	[CXL_MEM_COMMAND_ID_##_id] = {                                         \
33 	.info =	{                                                              \
34 			.id = CXL_MEM_COMMAND_ID_##_id,                        \
35 			.size_in = sin,                                        \
36 			.size_out = sout,                                      \
37 		},                                                             \
38 	.opcode = CXL_MBOX_OP_##_id,                                           \
39 	.flags = _flags,                                                       \
40 	}
41 
42 #define CXL_VARIABLE_PAYLOAD	~0U
43 /*
44  * This table defines the supported mailbox commands for the driver. This table
45  * is made up of a UAPI structure. Non-negative values as parameters in the
46  * table will be validated against the user's input. For example, if size_in is
47  * 0, and the user passed in 1, it is an error.
48  */
49 static struct cxl_mem_command cxl_mem_commands[CXL_MEM_COMMAND_ID_MAX] = {
50 	CXL_CMD(IDENTIFY, 0, 0x43, CXL_CMD_FLAG_FORCE_ENABLE),
51 #ifdef CONFIG_CXL_MEM_RAW_COMMANDS
52 	CXL_CMD(RAW, CXL_VARIABLE_PAYLOAD, CXL_VARIABLE_PAYLOAD, 0),
53 #endif
54 	CXL_CMD(GET_SUPPORTED_LOGS, 0, CXL_VARIABLE_PAYLOAD, CXL_CMD_FLAG_FORCE_ENABLE),
55 	CXL_CMD(GET_FW_INFO, 0, 0x50, 0),
56 	CXL_CMD(GET_PARTITION_INFO, 0, 0x20, 0),
57 	CXL_CMD(GET_LSA, 0x8, CXL_VARIABLE_PAYLOAD, 0),
58 	CXL_CMD(GET_HEALTH_INFO, 0, 0x12, 0),
59 	CXL_CMD(GET_LOG, 0x18, CXL_VARIABLE_PAYLOAD, CXL_CMD_FLAG_FORCE_ENABLE),
60 	CXL_CMD(GET_LOG_CAPS, 0x10, 0x4, 0),
61 	CXL_CMD(CLEAR_LOG, 0x10, 0, 0),
62 	CXL_CMD(GET_SUP_LOG_SUBLIST, 0x2, CXL_VARIABLE_PAYLOAD, 0),
63 	CXL_CMD(SET_PARTITION_INFO, 0x0a, 0, 0),
64 	CXL_CMD(SET_LSA, CXL_VARIABLE_PAYLOAD, 0, 0),
65 	CXL_CMD(GET_ALERT_CONFIG, 0, 0x10, 0),
66 	CXL_CMD(SET_ALERT_CONFIG, 0xc, 0, 0),
67 	CXL_CMD(GET_SHUTDOWN_STATE, 0, 0x1, 0),
68 	CXL_CMD(SET_SHUTDOWN_STATE, 0x1, 0, 0),
69 	CXL_CMD(GET_SCAN_MEDIA_CAPS, 0x10, 0x4, 0),
70 	CXL_CMD(GET_TIMESTAMP, 0, 0x8, 0),
71 };
72 
73 /*
74  * Commands that RAW doesn't permit. The rationale for each:
75  *
76  * CXL_MBOX_OP_ACTIVATE_FW: Firmware activation requires adjustment /
77  * coordination of transaction timeout values at the root bridge level.
78  *
79  * CXL_MBOX_OP_SET_PARTITION_INFO: The device memory map may change live
80  * and needs to be coordinated with HDM updates.
81  *
82  * CXL_MBOX_OP_SET_LSA: The label storage area may be cached by the
83  * driver and any writes from userspace invalidates those contents.
84  *
85  * CXL_MBOX_OP_SET_SHUTDOWN_STATE: Set shutdown state assumes no writes
86  * to the device after it is marked clean, userspace can not make that
87  * assertion.
88  *
89  * CXL_MBOX_OP_[GET_]SCAN_MEDIA: The kernel provides a native error list that
90  * is kept up to date with patrol notifications and error management.
91  *
92  * CXL_MBOX_OP_[GET_,INJECT_,CLEAR_]POISON: These commands require kernel
93  * driver orchestration for safety.
94  */
95 static u16 cxl_disabled_raw_commands[] = {
96 	CXL_MBOX_OP_ACTIVATE_FW,
97 	CXL_MBOX_OP_SET_PARTITION_INFO,
98 	CXL_MBOX_OP_SET_LSA,
99 	CXL_MBOX_OP_SET_SHUTDOWN_STATE,
100 	CXL_MBOX_OP_SCAN_MEDIA,
101 	CXL_MBOX_OP_GET_SCAN_MEDIA,
102 	CXL_MBOX_OP_GET_POISON,
103 	CXL_MBOX_OP_INJECT_POISON,
104 	CXL_MBOX_OP_CLEAR_POISON,
105 };
106 
107 /*
108  * Command sets that RAW doesn't permit. All opcodes in this set are
109  * disabled because they pass plain text security payloads over the
110  * user/kernel boundary. This functionality is intended to be wrapped
111  * behind the keys ABI which allows for encrypted payloads in the UAPI
112  */
113 static u8 security_command_sets[] = {
114 	0x44, /* Sanitize */
115 	0x45, /* Persistent Memory Data-at-rest Security */
116 	0x46, /* Security Passthrough */
117 };
118 
cxl_is_security_command(u16 opcode)119 static bool cxl_is_security_command(u16 opcode)
120 {
121 	int i;
122 
123 	for (i = 0; i < ARRAY_SIZE(security_command_sets); i++)
124 		if (security_command_sets[i] == (opcode >> 8))
125 			return true;
126 	return false;
127 }
128 
cxl_set_security_cmd_enabled(struct cxl_security_state * security,u16 opcode)129 static void cxl_set_security_cmd_enabled(struct cxl_security_state *security,
130 					 u16 opcode)
131 {
132 	switch (opcode) {
133 	case CXL_MBOX_OP_SANITIZE:
134 		set_bit(CXL_SEC_ENABLED_SANITIZE, security->enabled_cmds);
135 		break;
136 	case CXL_MBOX_OP_SECURE_ERASE:
137 		set_bit(CXL_SEC_ENABLED_SECURE_ERASE,
138 			security->enabled_cmds);
139 		break;
140 	case CXL_MBOX_OP_GET_SECURITY_STATE:
141 		set_bit(CXL_SEC_ENABLED_GET_SECURITY_STATE,
142 			security->enabled_cmds);
143 		break;
144 	case CXL_MBOX_OP_SET_PASSPHRASE:
145 		set_bit(CXL_SEC_ENABLED_SET_PASSPHRASE,
146 			security->enabled_cmds);
147 		break;
148 	case CXL_MBOX_OP_DISABLE_PASSPHRASE:
149 		set_bit(CXL_SEC_ENABLED_DISABLE_PASSPHRASE,
150 			security->enabled_cmds);
151 		break;
152 	case CXL_MBOX_OP_UNLOCK:
153 		set_bit(CXL_SEC_ENABLED_UNLOCK, security->enabled_cmds);
154 		break;
155 	case CXL_MBOX_OP_FREEZE_SECURITY:
156 		set_bit(CXL_SEC_ENABLED_FREEZE_SECURITY,
157 			security->enabled_cmds);
158 		break;
159 	case CXL_MBOX_OP_PASSPHRASE_SECURE_ERASE:
160 		set_bit(CXL_SEC_ENABLED_PASSPHRASE_SECURE_ERASE,
161 			security->enabled_cmds);
162 		break;
163 	default:
164 		break;
165 	}
166 }
167 
cxl_is_poison_command(u16 opcode)168 static bool cxl_is_poison_command(u16 opcode)
169 {
170 #define CXL_MBOX_OP_POISON_CMDS 0x43
171 
172 	if ((opcode >> 8) == CXL_MBOX_OP_POISON_CMDS)
173 		return true;
174 
175 	return false;
176 }
177 
cxl_set_poison_cmd_enabled(struct cxl_poison_state * poison,u16 opcode)178 static void cxl_set_poison_cmd_enabled(struct cxl_poison_state *poison,
179 				       u16 opcode)
180 {
181 	switch (opcode) {
182 	case CXL_MBOX_OP_GET_POISON:
183 		set_bit(CXL_POISON_ENABLED_LIST, poison->enabled_cmds);
184 		break;
185 	case CXL_MBOX_OP_INJECT_POISON:
186 		set_bit(CXL_POISON_ENABLED_INJECT, poison->enabled_cmds);
187 		break;
188 	case CXL_MBOX_OP_CLEAR_POISON:
189 		set_bit(CXL_POISON_ENABLED_CLEAR, poison->enabled_cmds);
190 		break;
191 	case CXL_MBOX_OP_GET_SCAN_MEDIA_CAPS:
192 		set_bit(CXL_POISON_ENABLED_SCAN_CAPS, poison->enabled_cmds);
193 		break;
194 	case CXL_MBOX_OP_SCAN_MEDIA:
195 		set_bit(CXL_POISON_ENABLED_SCAN_MEDIA, poison->enabled_cmds);
196 		break;
197 	case CXL_MBOX_OP_GET_SCAN_MEDIA:
198 		set_bit(CXL_POISON_ENABLED_SCAN_RESULTS, poison->enabled_cmds);
199 		break;
200 	default:
201 		break;
202 	}
203 }
204 
cxl_mem_find_command(u16 opcode)205 static struct cxl_mem_command *cxl_mem_find_command(u16 opcode)
206 {
207 	struct cxl_mem_command *c;
208 
209 	cxl_for_each_cmd(c)
210 		if (c->opcode == opcode)
211 			return c;
212 
213 	return NULL;
214 }
215 
cxl_mem_opcode_to_name(u16 opcode)216 static const char *cxl_mem_opcode_to_name(u16 opcode)
217 {
218 	struct cxl_mem_command *c;
219 
220 	c = cxl_mem_find_command(opcode);
221 	if (!c)
222 		return NULL;
223 
224 	return cxl_command_names[c->info.id].name;
225 }
226 
227 /**
228  * cxl_internal_send_cmd() - Kernel internal interface to send a mailbox command
229  * @cxl_mbox: CXL mailbox context
230  * @mbox_cmd: initialized command to execute
231  *
232  * Context: Any context.
233  * Return:
234  *  * %>=0	- Number of bytes returned in @out.
235  *  * %-E2BIG	- Payload is too large for hardware.
236  *  * %-EBUSY	- Couldn't acquire exclusive mailbox access.
237  *  * %-EFAULT	- Hardware error occurred.
238  *  * %-ENXIO	- Command completed, but device reported an error.
239  *  * %-EIO	- Unexpected output size.
240  *
241  * Mailbox commands may execute successfully yet the device itself reported an
242  * error. While this distinction can be useful for commands from userspace, the
243  * kernel will only be able to use results when both are successful.
244  */
cxl_internal_send_cmd(struct cxl_mailbox * cxl_mbox,struct cxl_mbox_cmd * mbox_cmd)245 int cxl_internal_send_cmd(struct cxl_mailbox *cxl_mbox,
246 			  struct cxl_mbox_cmd *mbox_cmd)
247 {
248 	size_t out_size, min_out;
249 	int rc;
250 
251 	if (mbox_cmd->size_in > cxl_mbox->payload_size ||
252 	    mbox_cmd->size_out > cxl_mbox->payload_size)
253 		return -E2BIG;
254 
255 	out_size = mbox_cmd->size_out;
256 	min_out = mbox_cmd->min_out;
257 	rc = cxl_mbox->mbox_send(cxl_mbox, mbox_cmd);
258 	/*
259 	 * EIO is reserved for a payload size mismatch and mbox_send()
260 	 * may not return this error.
261 	 */
262 	if (WARN_ONCE(rc == -EIO, "Bad return code: -EIO"))
263 		return -ENXIO;
264 	if (rc)
265 		return rc;
266 
267 	if (mbox_cmd->return_code != CXL_MBOX_CMD_RC_SUCCESS &&
268 	    mbox_cmd->return_code != CXL_MBOX_CMD_RC_BACKGROUND)
269 		return cxl_mbox_cmd_rc2errno(mbox_cmd);
270 
271 	if (!out_size)
272 		return 0;
273 
274 	/*
275 	 * Variable sized output needs to at least satisfy the caller's
276 	 * minimum if not the fully requested size.
277 	 */
278 	if (min_out == 0)
279 		min_out = out_size;
280 
281 	if (mbox_cmd->size_out < min_out)
282 		return -EIO;
283 	return 0;
284 }
285 EXPORT_SYMBOL_NS_GPL(cxl_internal_send_cmd, "CXL");
286 
cxl_mem_raw_command_allowed(u16 opcode)287 static bool cxl_mem_raw_command_allowed(u16 opcode)
288 {
289 	int i;
290 
291 	if (!IS_ENABLED(CONFIG_CXL_MEM_RAW_COMMANDS))
292 		return false;
293 
294 	if (security_locked_down(LOCKDOWN_PCI_ACCESS))
295 		return false;
296 
297 	if (cxl_raw_allow_all)
298 		return true;
299 
300 	if (cxl_is_security_command(opcode))
301 		return false;
302 
303 	for (i = 0; i < ARRAY_SIZE(cxl_disabled_raw_commands); i++)
304 		if (cxl_disabled_raw_commands[i] == opcode)
305 			return false;
306 
307 	return true;
308 }
309 
310 /**
311  * cxl_payload_from_user_allowed() - Check contents of in_payload.
312  * @opcode: The mailbox command opcode.
313  * @payload_in: Pointer to the input payload passed in from user space.
314  * @in_size: Size of @payload_in in bytes.
315  *
316  * Return:
317  *  * true	- payload_in passes check for @opcode.
318  *  * false	- payload_in contains invalid or unsupported values.
319  *
320  * The driver may inspect payload contents before sending a mailbox
321  * command from user space to the device. The intent is to reject
322  * commands with input payloads that are known to be unsafe. This
323  * check is not intended to replace the users careful selection of
324  * mailbox command parameters and makes no guarantee that the user
325  * command will succeed, nor that it is appropriate.
326  *
327  * The specific checks are determined by the opcode.
328  */
cxl_payload_from_user_allowed(u16 opcode,void * payload_in,size_t in_size)329 static bool cxl_payload_from_user_allowed(u16 opcode, void *payload_in,
330 					  size_t in_size)
331 {
332 	switch (opcode) {
333 	case CXL_MBOX_OP_SET_PARTITION_INFO: {
334 		struct cxl_mbox_set_partition_info *pi = payload_in;
335 
336 		if (in_size < sizeof(*pi))
337 			return false;
338 		if (pi->flags & CXL_SET_PARTITION_IMMEDIATE_FLAG)
339 			return false;
340 		break;
341 	}
342 	case CXL_MBOX_OP_CLEAR_LOG: {
343 		const uuid_t *uuid = (uuid_t *)payload_in;
344 
345 		if (in_size < sizeof(uuid_t))
346 			return false;
347 		/*
348 		 * Restrict the ‘Clear log’ action to only apply to
349 		 * Vendor debug logs.
350 		 */
351 		return uuid_equal(uuid, &DEFINE_CXL_VENDOR_DEBUG_UUID);
352 	}
353 	default:
354 		break;
355 	}
356 	return true;
357 }
358 
cxl_mbox_cmd_ctor(struct cxl_mbox_cmd * mbox_cmd,struct cxl_mailbox * cxl_mbox,u16 opcode,size_t in_size,size_t out_size,u64 in_payload)359 static int cxl_mbox_cmd_ctor(struct cxl_mbox_cmd *mbox_cmd,
360 			     struct cxl_mailbox *cxl_mbox, u16 opcode,
361 			     size_t in_size, size_t out_size, u64 in_payload)
362 {
363 	*mbox_cmd = (struct cxl_mbox_cmd) {
364 		.opcode = opcode,
365 		.size_in = in_size,
366 	};
367 
368 	if (in_size) {
369 		mbox_cmd->payload_in = vmemdup_user(u64_to_user_ptr(in_payload),
370 						    in_size);
371 		if (IS_ERR(mbox_cmd->payload_in))
372 			return PTR_ERR(mbox_cmd->payload_in);
373 
374 		if (!cxl_payload_from_user_allowed(opcode, mbox_cmd->payload_in,
375 						  in_size)) {
376 			dev_dbg(cxl_mbox->host, "%s: input payload not allowed\n",
377 				cxl_mem_opcode_to_name(opcode));
378 			kvfree(mbox_cmd->payload_in);
379 			return -EBUSY;
380 		}
381 	}
382 
383 	/* Prepare to handle a full payload for variable sized output */
384 	if (out_size == CXL_VARIABLE_PAYLOAD)
385 		mbox_cmd->size_out = cxl_mbox->payload_size;
386 	else
387 		mbox_cmd->size_out = out_size;
388 
389 	if (mbox_cmd->size_out) {
390 		mbox_cmd->payload_out = kvzalloc(mbox_cmd->size_out, GFP_KERNEL);
391 		if (!mbox_cmd->payload_out) {
392 			kvfree(mbox_cmd->payload_in);
393 			return -ENOMEM;
394 		}
395 	}
396 	return 0;
397 }
398 
cxl_mbox_cmd_dtor(struct cxl_mbox_cmd * mbox)399 static void cxl_mbox_cmd_dtor(struct cxl_mbox_cmd *mbox)
400 {
401 	kvfree(mbox->payload_in);
402 	kvfree(mbox->payload_out);
403 }
404 
cxl_to_mem_cmd_raw(struct cxl_mem_command * mem_cmd,const struct cxl_send_command * send_cmd,struct cxl_mailbox * cxl_mbox)405 static int cxl_to_mem_cmd_raw(struct cxl_mem_command *mem_cmd,
406 			      const struct cxl_send_command *send_cmd,
407 			      struct cxl_mailbox *cxl_mbox)
408 {
409 	if (send_cmd->raw.rsvd)
410 		return -EINVAL;
411 
412 	/*
413 	 * Unlike supported commands, the output size of RAW commands
414 	 * gets passed along without further checking, so it must be
415 	 * validated here.
416 	 */
417 	if (send_cmd->out.size > cxl_mbox->payload_size)
418 		return -EINVAL;
419 
420 	if (!cxl_mem_raw_command_allowed(send_cmd->raw.opcode))
421 		return -EPERM;
422 
423 	dev_WARN_ONCE(cxl_mbox->host, true, "raw command path used\n");
424 
425 	*mem_cmd = (struct cxl_mem_command) {
426 		.info = {
427 			.id = CXL_MEM_COMMAND_ID_RAW,
428 			.size_in = send_cmd->in.size,
429 			.size_out = send_cmd->out.size,
430 		},
431 		.opcode = send_cmd->raw.opcode
432 	};
433 
434 	return 0;
435 }
436 
cxl_to_mem_cmd(struct cxl_mem_command * mem_cmd,const struct cxl_send_command * send_cmd,struct cxl_mailbox * cxl_mbox)437 static int cxl_to_mem_cmd(struct cxl_mem_command *mem_cmd,
438 			  const struct cxl_send_command *send_cmd,
439 			  struct cxl_mailbox *cxl_mbox)
440 {
441 	struct cxl_mem_command *c = &cxl_mem_commands[send_cmd->id];
442 	const struct cxl_command_info *info = &c->info;
443 
444 	if (send_cmd->flags & ~CXL_MEM_COMMAND_FLAG_MASK)
445 		return -EINVAL;
446 
447 	if (send_cmd->rsvd)
448 		return -EINVAL;
449 
450 	if (send_cmd->in.rsvd || send_cmd->out.rsvd)
451 		return -EINVAL;
452 
453 	/* Check that the command is enabled for hardware */
454 	if (!test_bit(info->id, cxl_mbox->enabled_cmds))
455 		return -ENOTTY;
456 
457 	/* Check that the command is not claimed for exclusive kernel use */
458 	if (test_bit(info->id, cxl_mbox->exclusive_cmds))
459 		return -EBUSY;
460 
461 	/* Check the input buffer is the expected size */
462 	if ((info->size_in != CXL_VARIABLE_PAYLOAD) &&
463 	    (info->size_in != send_cmd->in.size))
464 		return -ENOMEM;
465 
466 	/* Check the output buffer is at least large enough */
467 	if ((info->size_out != CXL_VARIABLE_PAYLOAD) &&
468 	    (send_cmd->out.size < info->size_out))
469 		return -ENOMEM;
470 
471 	*mem_cmd = (struct cxl_mem_command) {
472 		.info = {
473 			.id = info->id,
474 			.flags = info->flags,
475 			.size_in = send_cmd->in.size,
476 			.size_out = send_cmd->out.size,
477 		},
478 		.opcode = c->opcode
479 	};
480 
481 	return 0;
482 }
483 
484 /**
485  * cxl_validate_cmd_from_user() - Check fields for CXL_MEM_SEND_COMMAND.
486  * @mbox_cmd: Sanitized and populated &struct cxl_mbox_cmd.
487  * @cxl_mbox: CXL mailbox context
488  * @send_cmd: &struct cxl_send_command copied in from userspace.
489  *
490  * Return:
491  *  * %0	- @out_cmd is ready to send.
492  *  * %-ENOTTY	- Invalid command specified.
493  *  * %-EINVAL	- Reserved fields or invalid values were used.
494  *  * %-ENOMEM	- Input or output buffer wasn't sized properly.
495  *  * %-EPERM	- Attempted to use a protected command.
496  *  * %-EBUSY	- Kernel has claimed exclusive access to this opcode
497  *
498  * The result of this command is a fully validated command in @mbox_cmd that is
499  * safe to send to the hardware.
500  */
cxl_validate_cmd_from_user(struct cxl_mbox_cmd * mbox_cmd,struct cxl_mailbox * cxl_mbox,const struct cxl_send_command * send_cmd)501 static int cxl_validate_cmd_from_user(struct cxl_mbox_cmd *mbox_cmd,
502 				      struct cxl_mailbox *cxl_mbox,
503 				      const struct cxl_send_command *send_cmd)
504 {
505 	struct cxl_mem_command mem_cmd;
506 	int rc;
507 
508 	if (send_cmd->id == 0 || send_cmd->id >= CXL_MEM_COMMAND_ID_MAX)
509 		return -ENOTTY;
510 
511 	/*
512 	 * The user can never specify an input payload larger than what hardware
513 	 * supports, but output can be arbitrarily large (simply write out as
514 	 * much data as the hardware provides).
515 	 */
516 	if (send_cmd->in.size > cxl_mbox->payload_size)
517 		return -EINVAL;
518 
519 	/* Sanitize and construct a cxl_mem_command */
520 	if (send_cmd->id == CXL_MEM_COMMAND_ID_RAW)
521 		rc = cxl_to_mem_cmd_raw(&mem_cmd, send_cmd, cxl_mbox);
522 	else
523 		rc = cxl_to_mem_cmd(&mem_cmd, send_cmd, cxl_mbox);
524 
525 	if (rc)
526 		return rc;
527 
528 	/* Sanitize and construct a cxl_mbox_cmd */
529 	return cxl_mbox_cmd_ctor(mbox_cmd, cxl_mbox, mem_cmd.opcode,
530 				 mem_cmd.info.size_in, mem_cmd.info.size_out,
531 				 send_cmd->in.payload);
532 }
533 
cxl_query_cmd(struct cxl_mailbox * cxl_mbox,struct cxl_mem_query_commands __user * q)534 int cxl_query_cmd(struct cxl_mailbox *cxl_mbox,
535 		  struct cxl_mem_query_commands __user *q)
536 {
537 	struct device *dev = cxl_mbox->host;
538 	struct cxl_mem_command *cmd;
539 	u32 n_commands;
540 	int j = 0;
541 
542 	dev_dbg(dev, "Query IOCTL\n");
543 
544 	if (get_user(n_commands, &q->n_commands))
545 		return -EFAULT;
546 
547 	/* returns the total number if 0 elements are requested. */
548 	if (n_commands == 0)
549 		return put_user(ARRAY_SIZE(cxl_mem_commands), &q->n_commands);
550 
551 	/*
552 	 * otherwise, return min(n_commands, total commands) cxl_command_info
553 	 * structures.
554 	 */
555 	cxl_for_each_cmd(cmd) {
556 		struct cxl_command_info info = cmd->info;
557 
558 		if (test_bit(info.id, cxl_mbox->enabled_cmds))
559 			info.flags |= CXL_MEM_COMMAND_FLAG_ENABLED;
560 		if (test_bit(info.id, cxl_mbox->exclusive_cmds))
561 			info.flags |= CXL_MEM_COMMAND_FLAG_EXCLUSIVE;
562 
563 		if (copy_to_user(&q->commands[j++], &info, sizeof(info)))
564 			return -EFAULT;
565 
566 		if (j == n_commands)
567 			break;
568 	}
569 
570 	return 0;
571 }
572 
573 /**
574  * handle_mailbox_cmd_from_user() - Dispatch a mailbox command for userspace.
575  * @cxl_mbox: The mailbox context for the operation.
576  * @mbox_cmd: The validated mailbox command.
577  * @out_payload: Pointer to userspace's output payload.
578  * @size_out: (Input) Max payload size to copy out.
579  *            (Output) Payload size hardware generated.
580  * @retval: Hardware generated return code from the operation.
581  *
582  * Return:
583  *  * %0	- Mailbox transaction succeeded. This implies the mailbox
584  *		  protocol completed successfully not that the operation itself
585  *		  was successful.
586  *  * %-ENOMEM  - Couldn't allocate a bounce buffer.
587  *  * %-EFAULT	- Something happened with copy_to/from_user.
588  *  * %-EINTR	- Mailbox acquisition interrupted.
589  *  * %-EXXX	- Transaction level failures.
590  *
591  * Dispatches a mailbox command on behalf of a userspace request.
592  * The output payload is copied to userspace.
593  *
594  * See cxl_send_cmd().
595  */
handle_mailbox_cmd_from_user(struct cxl_mailbox * cxl_mbox,struct cxl_mbox_cmd * mbox_cmd,u64 out_payload,s32 * size_out,u32 * retval)596 static int handle_mailbox_cmd_from_user(struct cxl_mailbox *cxl_mbox,
597 					struct cxl_mbox_cmd *mbox_cmd,
598 					u64 out_payload, s32 *size_out,
599 					u32 *retval)
600 {
601 	struct device *dev = cxl_mbox->host;
602 	int rc;
603 
604 	dev_dbg(dev,
605 		"Submitting %s command for user\n"
606 		"\topcode: %x\n"
607 		"\tsize: %zx\n",
608 		cxl_mem_opcode_to_name(mbox_cmd->opcode),
609 		mbox_cmd->opcode, mbox_cmd->size_in);
610 
611 	rc = cxl_mbox->mbox_send(cxl_mbox, mbox_cmd);
612 	if (rc)
613 		goto out;
614 
615 	/*
616 	 * @size_out contains the max size that's allowed to be written back out
617 	 * to userspace. While the payload may have written more output than
618 	 * this it will have to be ignored.
619 	 */
620 	if (mbox_cmd->size_out) {
621 		dev_WARN_ONCE(dev, mbox_cmd->size_out > *size_out,
622 			      "Invalid return size\n");
623 		if (copy_to_user(u64_to_user_ptr(out_payload),
624 				 mbox_cmd->payload_out, mbox_cmd->size_out)) {
625 			rc = -EFAULT;
626 			goto out;
627 		}
628 	}
629 
630 	*size_out = mbox_cmd->size_out;
631 	*retval = mbox_cmd->return_code;
632 
633 out:
634 	cxl_mbox_cmd_dtor(mbox_cmd);
635 	return rc;
636 }
637 
cxl_send_cmd(struct cxl_mailbox * cxl_mbox,struct cxl_send_command __user * s)638 int cxl_send_cmd(struct cxl_mailbox *cxl_mbox, struct cxl_send_command __user *s)
639 {
640 	struct device *dev = cxl_mbox->host;
641 	struct cxl_send_command send;
642 	struct cxl_mbox_cmd mbox_cmd;
643 	int rc;
644 
645 	dev_dbg(dev, "Send IOCTL\n");
646 
647 	if (copy_from_user(&send, s, sizeof(send)))
648 		return -EFAULT;
649 
650 	rc = cxl_validate_cmd_from_user(&mbox_cmd, cxl_mbox, &send);
651 	if (rc)
652 		return rc;
653 
654 	rc = handle_mailbox_cmd_from_user(cxl_mbox, &mbox_cmd, send.out.payload,
655 					  &send.out.size, &send.retval);
656 	if (rc)
657 		return rc;
658 
659 	if (copy_to_user(s, &send, sizeof(send)))
660 		return -EFAULT;
661 
662 	return 0;
663 }
664 
cxl_xfer_log(struct cxl_memdev_state * mds,uuid_t * uuid,u32 * size,u8 * out)665 static int cxl_xfer_log(struct cxl_memdev_state *mds, uuid_t *uuid,
666 			u32 *size, u8 *out)
667 {
668 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
669 	u32 remaining = *size;
670 	u32 offset = 0;
671 
672 	while (remaining) {
673 		u32 xfer_size = min_t(u32, remaining, cxl_mbox->payload_size);
674 		struct cxl_mbox_cmd mbox_cmd;
675 		struct cxl_mbox_get_log log;
676 		int rc;
677 
678 		log = (struct cxl_mbox_get_log) {
679 			.uuid = *uuid,
680 			.offset = cpu_to_le32(offset),
681 			.length = cpu_to_le32(xfer_size),
682 		};
683 
684 		mbox_cmd = (struct cxl_mbox_cmd) {
685 			.opcode = CXL_MBOX_OP_GET_LOG,
686 			.size_in = sizeof(log),
687 			.payload_in = &log,
688 			.size_out = xfer_size,
689 			.payload_out = out,
690 		};
691 
692 		rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
693 
694 		/*
695 		 * The output payload length that indicates the number
696 		 * of valid bytes can be smaller than the Log buffer
697 		 * size.
698 		 */
699 		if (rc == -EIO && mbox_cmd.size_out < xfer_size) {
700 			offset += mbox_cmd.size_out;
701 			break;
702 		}
703 
704 		if (rc < 0)
705 			return rc;
706 
707 		out += xfer_size;
708 		remaining -= xfer_size;
709 		offset += xfer_size;
710 	}
711 
712 	*size = offset;
713 
714 	return 0;
715 }
716 
check_features_opcodes(u16 opcode,int * ro_cmds,int * wr_cmds)717 static int check_features_opcodes(u16 opcode, int *ro_cmds, int *wr_cmds)
718 {
719 	switch (opcode) {
720 	case CXL_MBOX_OP_GET_SUPPORTED_FEATURES:
721 	case CXL_MBOX_OP_GET_FEATURE:
722 		(*ro_cmds)++;
723 		return 1;
724 	case CXL_MBOX_OP_SET_FEATURE:
725 		(*wr_cmds)++;
726 		return 1;
727 	default:
728 		return 0;
729 	}
730 }
731 
732 /* 'Get Supported Features' and 'Get Feature' */
733 #define MAX_FEATURES_READ_CMDS	2
set_features_cap(struct cxl_mailbox * cxl_mbox,int ro_cmds,int wr_cmds)734 static void set_features_cap(struct cxl_mailbox *cxl_mbox,
735 			     int ro_cmds, int wr_cmds)
736 {
737 	/* Setting up Features capability while walking the CEL */
738 	if (ro_cmds == MAX_FEATURES_READ_CMDS) {
739 		if (wr_cmds)
740 			cxl_mbox->feat_cap = CXL_FEATURES_RW;
741 		else
742 			cxl_mbox->feat_cap = CXL_FEATURES_RO;
743 	}
744 }
745 
746 /**
747  * cxl_walk_cel() - Walk through the Command Effects Log.
748  * @mds: The driver data for the operation
749  * @size: Length of the Command Effects Log.
750  * @cel: CEL
751  *
752  * Iterate over each entry in the CEL and determine if the driver supports the
753  * command. If so, the command is enabled for the device and can be used later.
754  */
cxl_walk_cel(struct cxl_memdev_state * mds,size_t size,u8 * cel)755 static void cxl_walk_cel(struct cxl_memdev_state *mds, size_t size, u8 *cel)
756 {
757 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
758 	struct cxl_cel_entry *cel_entry;
759 	const int cel_entries = size / sizeof(*cel_entry);
760 	struct device *dev = mds->cxlds.dev;
761 	int i, ro_cmds = 0, wr_cmds = 0;
762 
763 	cel_entry = (struct cxl_cel_entry *) cel;
764 
765 	for (i = 0; i < cel_entries; i++) {
766 		u16 opcode = le16_to_cpu(cel_entry[i].opcode);
767 		struct cxl_mem_command *cmd = cxl_mem_find_command(opcode);
768 		int enabled = 0;
769 
770 		if (cmd) {
771 			set_bit(cmd->info.id, cxl_mbox->enabled_cmds);
772 			enabled++;
773 		}
774 
775 		enabled += check_features_opcodes(opcode, &ro_cmds,
776 						  &wr_cmds);
777 
778 		if (cxl_is_poison_command(opcode)) {
779 			cxl_set_poison_cmd_enabled(&mds->poison, opcode);
780 			enabled++;
781 		}
782 
783 		if (cxl_is_security_command(opcode)) {
784 			cxl_set_security_cmd_enabled(&mds->security, opcode);
785 			enabled++;
786 		}
787 
788 		dev_dbg(dev, "Opcode 0x%04x %s\n", opcode,
789 			enabled ? "enabled" : "unsupported by driver");
790 	}
791 
792 	set_features_cap(cxl_mbox, ro_cmds, wr_cmds);
793 }
794 
cxl_get_gsl(struct cxl_memdev_state * mds)795 static struct cxl_mbox_get_supported_logs *cxl_get_gsl(struct cxl_memdev_state *mds)
796 {
797 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
798 	struct cxl_mbox_get_supported_logs *ret;
799 	struct cxl_mbox_cmd mbox_cmd;
800 	int rc;
801 
802 	ret = kvmalloc(cxl_mbox->payload_size, GFP_KERNEL);
803 	if (!ret)
804 		return ERR_PTR(-ENOMEM);
805 
806 	mbox_cmd = (struct cxl_mbox_cmd) {
807 		.opcode = CXL_MBOX_OP_GET_SUPPORTED_LOGS,
808 		.size_out = cxl_mbox->payload_size,
809 		.payload_out = ret,
810 		/* At least the record number field must be valid */
811 		.min_out = 2,
812 	};
813 	rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
814 	if (rc < 0) {
815 		kvfree(ret);
816 		return ERR_PTR(rc);
817 	}
818 
819 
820 	return ret;
821 }
822 
823 enum {
824 	CEL_UUID,
825 	VENDOR_DEBUG_UUID,
826 };
827 
828 /* See CXL 2.0 Table 170. Get Log Input Payload */
829 static const uuid_t log_uuid[] = {
830 	[CEL_UUID] = DEFINE_CXL_CEL_UUID,
831 	[VENDOR_DEBUG_UUID] = DEFINE_CXL_VENDOR_DEBUG_UUID,
832 };
833 
834 /**
835  * cxl_enumerate_cmds() - Enumerate commands for a device.
836  * @mds: The driver data for the operation
837  *
838  * Returns 0 if enumerate completed successfully.
839  *
840  * CXL devices have optional support for certain commands. This function will
841  * determine the set of supported commands for the hardware and update the
842  * enabled_cmds bitmap in the @mds.
843  */
cxl_enumerate_cmds(struct cxl_memdev_state * mds)844 int cxl_enumerate_cmds(struct cxl_memdev_state *mds)
845 {
846 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
847 	struct cxl_mbox_get_supported_logs *gsl;
848 	struct device *dev = mds->cxlds.dev;
849 	struct cxl_mem_command *cmd;
850 	int i, rc;
851 
852 	gsl = cxl_get_gsl(mds);
853 	if (IS_ERR(gsl))
854 		return PTR_ERR(gsl);
855 
856 	rc = -ENOENT;
857 	for (i = 0; i < le16_to_cpu(gsl->entries); i++) {
858 		u32 size = le32_to_cpu(gsl->entry[i].size);
859 		uuid_t uuid = gsl->entry[i].uuid;
860 		u8 *log;
861 
862 		dev_dbg(dev, "Found LOG type %pU of size %d", &uuid, size);
863 
864 		if (!uuid_equal(&uuid, &log_uuid[CEL_UUID]))
865 			continue;
866 
867 		log = kvmalloc(size, GFP_KERNEL);
868 		if (!log) {
869 			rc = -ENOMEM;
870 			goto out;
871 		}
872 
873 		rc = cxl_xfer_log(mds, &uuid, &size, log);
874 		if (rc) {
875 			kvfree(log);
876 			goto out;
877 		}
878 
879 		cxl_walk_cel(mds, size, log);
880 		kvfree(log);
881 
882 		/* In case CEL was bogus, enable some default commands. */
883 		cxl_for_each_cmd(cmd)
884 			if (cmd->flags & CXL_CMD_FLAG_FORCE_ENABLE)
885 				set_bit(cmd->info.id, cxl_mbox->enabled_cmds);
886 
887 		/* Found the required CEL */
888 		rc = 0;
889 	}
890 out:
891 	kvfree(gsl);
892 	return rc;
893 }
894 EXPORT_SYMBOL_NS_GPL(cxl_enumerate_cmds, "CXL");
895 
cxl_event_trace_record(const struct cxl_memdev * cxlmd,enum cxl_event_log_type type,enum cxl_event_type event_type,const uuid_t * uuid,union cxl_event * evt)896 void cxl_event_trace_record(const struct cxl_memdev *cxlmd,
897 			    enum cxl_event_log_type type,
898 			    enum cxl_event_type event_type,
899 			    const uuid_t *uuid, union cxl_event *evt)
900 {
901 	if (event_type == CXL_CPER_EVENT_MEM_MODULE) {
902 		trace_cxl_memory_module(cxlmd, type, &evt->mem_module);
903 		return;
904 	}
905 	if (event_type == CXL_CPER_EVENT_GENERIC) {
906 		trace_cxl_generic_event(cxlmd, type, uuid, &evt->generic);
907 		return;
908 	}
909 	if (event_type == CXL_CPER_EVENT_MEM_SPARING) {
910 		trace_cxl_memory_sparing(cxlmd, type, &evt->mem_sparing);
911 		return;
912 	}
913 
914 	if (trace_cxl_general_media_enabled() || trace_cxl_dram_enabled()) {
915 		u64 dpa, hpa = ULLONG_MAX, hpa_alias = ULLONG_MAX;
916 		struct cxl_region *cxlr;
917 
918 		/*
919 		 * These trace points are annotated with HPA and region
920 		 * translations. Take topology mutation locks and lookup
921 		 * { HPA, REGION } from { DPA, MEMDEV } in the event record.
922 		 */
923 		guard(rwsem_read)(&cxl_rwsem.region);
924 		guard(rwsem_read)(&cxl_rwsem.dpa);
925 
926 		dpa = le64_to_cpu(evt->media_hdr.phys_addr) & CXL_DPA_MASK;
927 		cxlr = cxl_dpa_to_region(cxlmd, dpa);
928 		if (cxlr) {
929 			u64 cache_size = cxlr->params.cache_size;
930 
931 			hpa = cxl_dpa_to_hpa(cxlr, cxlmd, dpa);
932 			if (cache_size)
933 				hpa_alias = hpa - cache_size;
934 		}
935 
936 		if (event_type == CXL_CPER_EVENT_GEN_MEDIA) {
937 			if (cxl_store_rec_gen_media((struct cxl_memdev *)cxlmd, evt))
938 				dev_dbg(&cxlmd->dev, "CXL store rec_gen_media failed\n");
939 
940 			if (evt->gen_media.media_hdr.descriptor &
941 			    CXL_GMER_EVT_DESC_THRESHOLD_EVENT)
942 				WARN_ON_ONCE((evt->gen_media.media_hdr.type &
943 					      CXL_GMER_MEM_EVT_TYPE_AP_CME_COUNTER_EXPIRE) &&
944 					     !get_unaligned_le24(evt->gen_media.cme_count));
945 			else
946 				WARN_ON_ONCE(evt->gen_media.media_hdr.type &
947 					     CXL_GMER_MEM_EVT_TYPE_AP_CME_COUNTER_EXPIRE);
948 
949 			trace_cxl_general_media(cxlmd, type, cxlr, hpa,
950 						hpa_alias, &evt->gen_media);
951 		} else if (event_type == CXL_CPER_EVENT_DRAM) {
952 			if (cxl_store_rec_dram((struct cxl_memdev *)cxlmd, evt))
953 				dev_dbg(&cxlmd->dev, "CXL store rec_dram failed\n");
954 
955 			if (evt->dram.media_hdr.descriptor &
956 			    CXL_GMER_EVT_DESC_THRESHOLD_EVENT)
957 				WARN_ON_ONCE((evt->dram.media_hdr.type &
958 					      CXL_DER_MEM_EVT_TYPE_AP_CME_COUNTER_EXPIRE) &&
959 					     !get_unaligned_le24(evt->dram.cvme_count));
960 			else
961 				WARN_ON_ONCE(evt->dram.media_hdr.type &
962 					     CXL_DER_MEM_EVT_TYPE_AP_CME_COUNTER_EXPIRE);
963 
964 			trace_cxl_dram(cxlmd, type, cxlr, hpa, hpa_alias,
965 				       &evt->dram);
966 		}
967 	}
968 }
969 EXPORT_SYMBOL_NS_GPL(cxl_event_trace_record, "CXL");
970 
__cxl_event_trace_record(const struct cxl_memdev * cxlmd,enum cxl_event_log_type type,struct cxl_event_record_raw * record)971 static void __cxl_event_trace_record(const struct cxl_memdev *cxlmd,
972 				     enum cxl_event_log_type type,
973 				     struct cxl_event_record_raw *record)
974 {
975 	enum cxl_event_type ev_type = CXL_CPER_EVENT_GENERIC;
976 	const uuid_t *uuid = &record->id;
977 
978 	if (uuid_equal(uuid, &CXL_EVENT_GEN_MEDIA_UUID))
979 		ev_type = CXL_CPER_EVENT_GEN_MEDIA;
980 	else if (uuid_equal(uuid, &CXL_EVENT_DRAM_UUID))
981 		ev_type = CXL_CPER_EVENT_DRAM;
982 	else if (uuid_equal(uuid, &CXL_EVENT_MEM_MODULE_UUID))
983 		ev_type = CXL_CPER_EVENT_MEM_MODULE;
984 	else if (uuid_equal(uuid, &CXL_EVENT_MEM_SPARING_UUID))
985 		ev_type = CXL_CPER_EVENT_MEM_SPARING;
986 
987 	cxl_event_trace_record(cxlmd, type, ev_type, uuid, &record->event);
988 }
989 
cxl_clear_event_record(struct cxl_memdev_state * mds,enum cxl_event_log_type log,struct cxl_get_event_payload * get_pl)990 static int cxl_clear_event_record(struct cxl_memdev_state *mds,
991 				  enum cxl_event_log_type log,
992 				  struct cxl_get_event_payload *get_pl)
993 {
994 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
995 	struct cxl_mbox_clear_event_payload *payload;
996 	u16 total = le16_to_cpu(get_pl->record_count);
997 	u8 max_handles = CXL_CLEAR_EVENT_MAX_HANDLES;
998 	size_t pl_size = struct_size(payload, handles, max_handles);
999 	struct cxl_mbox_cmd mbox_cmd;
1000 	u16 cnt;
1001 	int rc = 0;
1002 	int i;
1003 
1004 	/* Payload size may limit the max handles */
1005 	if (pl_size > cxl_mbox->payload_size) {
1006 		max_handles = (cxl_mbox->payload_size - sizeof(*payload)) /
1007 			      sizeof(__le16);
1008 		pl_size = struct_size(payload, handles, max_handles);
1009 	}
1010 
1011 	payload = kvzalloc(pl_size, GFP_KERNEL);
1012 	if (!payload)
1013 		return -ENOMEM;
1014 
1015 	*payload = (struct cxl_mbox_clear_event_payload) {
1016 		.event_log = log,
1017 	};
1018 
1019 	mbox_cmd = (struct cxl_mbox_cmd) {
1020 		.opcode = CXL_MBOX_OP_CLEAR_EVENT_RECORD,
1021 		.payload_in = payload,
1022 		.size_in = pl_size,
1023 	};
1024 
1025 	/*
1026 	 * Clear Event Records uses u8 for the handle cnt while Get Event
1027 	 * Record can return up to 0xffff records.
1028 	 */
1029 	i = 0;
1030 	for (cnt = 0; cnt < total; cnt++) {
1031 		struct cxl_event_record_raw *raw = &get_pl->records[cnt];
1032 		struct cxl_event_generic *gen = &raw->event.generic;
1033 
1034 		payload->handles[i++] = gen->hdr.handle;
1035 		dev_dbg(mds->cxlds.dev, "Event log '%d': Clearing %u\n", log,
1036 			le16_to_cpu(payload->handles[i - 1]));
1037 
1038 		if (i == max_handles) {
1039 			payload->nr_recs = i;
1040 			rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1041 			if (rc)
1042 				goto free_pl;
1043 			i = 0;
1044 		}
1045 	}
1046 
1047 	/* Clear what is left if any */
1048 	if (i) {
1049 		payload->nr_recs = i;
1050 		mbox_cmd.size_in = struct_size(payload, handles, i);
1051 		rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1052 		if (rc)
1053 			goto free_pl;
1054 	}
1055 
1056 free_pl:
1057 	kvfree(payload);
1058 	return rc;
1059 }
1060 
cxl_mem_get_records_log(struct cxl_memdev_state * mds,enum cxl_event_log_type type)1061 static void cxl_mem_get_records_log(struct cxl_memdev_state *mds,
1062 				    enum cxl_event_log_type type)
1063 {
1064 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1065 	struct cxl_memdev *cxlmd = mds->cxlds.cxlmd;
1066 	struct device *dev = mds->cxlds.dev;
1067 	struct cxl_get_event_payload *payload;
1068 	u8 log_type = type;
1069 	u16 nr_rec;
1070 
1071 	mutex_lock(&mds->event.log_lock);
1072 	payload = mds->event.buf;
1073 
1074 	do {
1075 		int rc, i;
1076 		struct cxl_mbox_cmd mbox_cmd = (struct cxl_mbox_cmd) {
1077 			.opcode = CXL_MBOX_OP_GET_EVENT_RECORD,
1078 			.payload_in = &log_type,
1079 			.size_in = sizeof(log_type),
1080 			.payload_out = payload,
1081 			.size_out = cxl_mbox->payload_size,
1082 			.min_out = struct_size(payload, records, 0),
1083 		};
1084 
1085 		rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1086 		if (rc) {
1087 			dev_err_ratelimited(dev,
1088 				"Event log '%d': Failed to query event records : %d",
1089 				type, rc);
1090 			break;
1091 		}
1092 
1093 		nr_rec = le16_to_cpu(payload->record_count);
1094 		if (!nr_rec)
1095 			break;
1096 
1097 		for (i = 0; i < nr_rec; i++)
1098 			__cxl_event_trace_record(cxlmd, type,
1099 						 &payload->records[i]);
1100 
1101 		if (payload->flags & CXL_GET_EVENT_FLAG_OVERFLOW)
1102 			trace_cxl_overflow(cxlmd, type, payload);
1103 
1104 		rc = cxl_clear_event_record(mds, type, payload);
1105 		if (rc) {
1106 			dev_err_ratelimited(dev,
1107 				"Event log '%d': Failed to clear events : %d",
1108 				type, rc);
1109 			break;
1110 		}
1111 	} while (nr_rec);
1112 
1113 	mutex_unlock(&mds->event.log_lock);
1114 }
1115 
1116 /**
1117  * cxl_mem_get_event_records - Get Event Records from the device
1118  * @mds: The driver data for the operation
1119  * @status: Event Status register value identifying which events are available.
1120  *
1121  * Retrieve all event records available on the device, report them as trace
1122  * events, and clear them.
1123  *
1124  * See CXL rev 3.0 @8.2.9.2.2 Get Event Records
1125  * See CXL rev 3.0 @8.2.9.2.3 Clear Event Records
1126  */
cxl_mem_get_event_records(struct cxl_memdev_state * mds,u32 status)1127 void cxl_mem_get_event_records(struct cxl_memdev_state *mds, u32 status)
1128 {
1129 	dev_dbg(mds->cxlds.dev, "Reading event logs: %x\n", status);
1130 
1131 	if (status & CXLDEV_EVENT_STATUS_FATAL)
1132 		cxl_mem_get_records_log(mds, CXL_EVENT_TYPE_FATAL);
1133 	if (status & CXLDEV_EVENT_STATUS_FAIL)
1134 		cxl_mem_get_records_log(mds, CXL_EVENT_TYPE_FAIL);
1135 	if (status & CXLDEV_EVENT_STATUS_WARN)
1136 		cxl_mem_get_records_log(mds, CXL_EVENT_TYPE_WARN);
1137 	if (status & CXLDEV_EVENT_STATUS_INFO)
1138 		cxl_mem_get_records_log(mds, CXL_EVENT_TYPE_INFO);
1139 }
1140 EXPORT_SYMBOL_NS_GPL(cxl_mem_get_event_records, "CXL");
1141 
1142 /**
1143  * cxl_mem_get_partition_info - Get partition info
1144  * @mds: The driver data for the operation
1145  *
1146  * Retrieve the current partition info for the device specified.  The active
1147  * values are the current capacity in bytes.  If not 0, the 'next' values are
1148  * the pending values, in bytes, which take affect on next cold reset.
1149  *
1150  * Return: 0 if no error: or the result of the mailbox command.
1151  *
1152  * See CXL @8.2.9.5.2.1 Get Partition Info
1153  */
cxl_mem_get_partition_info(struct cxl_memdev_state * mds)1154 static int cxl_mem_get_partition_info(struct cxl_memdev_state *mds)
1155 {
1156 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1157 	struct cxl_mbox_get_partition_info pi;
1158 	struct cxl_mbox_cmd mbox_cmd;
1159 	int rc;
1160 
1161 	mbox_cmd = (struct cxl_mbox_cmd) {
1162 		.opcode = CXL_MBOX_OP_GET_PARTITION_INFO,
1163 		.size_out = sizeof(pi),
1164 		.payload_out = &pi,
1165 	};
1166 	rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1167 	if (rc)
1168 		return rc;
1169 
1170 	mds->active_volatile_bytes =
1171 		le64_to_cpu(pi.active_volatile_cap) * CXL_CAPACITY_MULTIPLIER;
1172 	mds->active_persistent_bytes =
1173 		le64_to_cpu(pi.active_persistent_cap) * CXL_CAPACITY_MULTIPLIER;
1174 
1175 	return 0;
1176 }
1177 
1178 /**
1179  * cxl_dev_state_identify() - Send the IDENTIFY command to the device.
1180  * @mds: The driver data for the operation
1181  *
1182  * Return: 0 if identify was executed successfully or media not ready.
1183  *
1184  * This will dispatch the identify command to the device and on success populate
1185  * structures to be exported to sysfs.
1186  */
cxl_dev_state_identify(struct cxl_memdev_state * mds)1187 int cxl_dev_state_identify(struct cxl_memdev_state *mds)
1188 {
1189 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1190 	/* See CXL 2.0 Table 175 Identify Memory Device Output Payload */
1191 	struct cxl_mbox_identify id;
1192 	struct cxl_mbox_cmd mbox_cmd;
1193 	u32 val;
1194 	int rc;
1195 
1196 	if (!mds->cxlds.media_ready)
1197 		return 0;
1198 
1199 	mbox_cmd = (struct cxl_mbox_cmd) {
1200 		.opcode = CXL_MBOX_OP_IDENTIFY,
1201 		.size_out = sizeof(id),
1202 		.payload_out = &id,
1203 	};
1204 	rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1205 	if (rc < 0)
1206 		return rc;
1207 
1208 	mds->total_bytes =
1209 		le64_to_cpu(id.total_capacity) * CXL_CAPACITY_MULTIPLIER;
1210 	mds->volatile_only_bytes =
1211 		le64_to_cpu(id.volatile_capacity) * CXL_CAPACITY_MULTIPLIER;
1212 	mds->persistent_only_bytes =
1213 		le64_to_cpu(id.persistent_capacity) * CXL_CAPACITY_MULTIPLIER;
1214 	mds->partition_align_bytes =
1215 		le64_to_cpu(id.partition_align) * CXL_CAPACITY_MULTIPLIER;
1216 
1217 	mds->lsa_size = le32_to_cpu(id.lsa_size);
1218 	memcpy(mds->firmware_version, id.fw_revision,
1219 	       sizeof(id.fw_revision));
1220 
1221 	if (test_bit(CXL_POISON_ENABLED_LIST, mds->poison.enabled_cmds)) {
1222 		val = get_unaligned_le24(id.poison_list_max_mer);
1223 		mds->poison.max_errors = min_t(u32, val, CXL_POISON_LIST_MAX);
1224 	}
1225 
1226 	return 0;
1227 }
1228 EXPORT_SYMBOL_NS_GPL(cxl_dev_state_identify, "CXL");
1229 
__cxl_mem_sanitize(struct cxl_memdev_state * mds,u16 cmd)1230 static int __cxl_mem_sanitize(struct cxl_memdev_state *mds, u16 cmd)
1231 {
1232 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1233 	int rc;
1234 	u32 sec_out = 0;
1235 	struct cxl_get_security_output {
1236 		__le32 flags;
1237 	} out;
1238 	struct cxl_mbox_cmd sec_cmd = {
1239 		.opcode = CXL_MBOX_OP_GET_SECURITY_STATE,
1240 		.payload_out = &out,
1241 		.size_out = sizeof(out),
1242 	};
1243 	struct cxl_mbox_cmd mbox_cmd = { .opcode = cmd };
1244 
1245 	if (cmd != CXL_MBOX_OP_SANITIZE && cmd != CXL_MBOX_OP_SECURE_ERASE)
1246 		return -EINVAL;
1247 
1248 	rc = cxl_internal_send_cmd(cxl_mbox, &sec_cmd);
1249 	if (rc < 0) {
1250 		dev_err(cxl_mbox->host, "Failed to get security state : %d", rc);
1251 		return rc;
1252 	}
1253 
1254 	/*
1255 	 * Prior to using these commands, any security applied to
1256 	 * the user data areas of the device shall be DISABLED (or
1257 	 * UNLOCKED for secure erase case).
1258 	 */
1259 	sec_out = le32_to_cpu(out.flags);
1260 	if (sec_out & CXL_PMEM_SEC_STATE_USER_PASS_SET)
1261 		return -EINVAL;
1262 
1263 	if (cmd == CXL_MBOX_OP_SECURE_ERASE &&
1264 	    sec_out & CXL_PMEM_SEC_STATE_LOCKED)
1265 		return -EINVAL;
1266 
1267 	rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1268 	if (rc < 0) {
1269 		dev_err(cxl_mbox->host, "Failed to sanitize device : %d", rc);
1270 		return rc;
1271 	}
1272 
1273 	return 0;
1274 }
1275 
1276 
1277 /**
1278  * cxl_mem_sanitize() - Send a sanitization command to the device.
1279  * @cxlmd: The device for the operation
1280  * @cmd: The specific sanitization command opcode
1281  *
1282  * Return: 0 if the command was executed successfully, regardless of
1283  * whether or not the actual security operation is done in the background,
1284  * such as for the Sanitize case.
1285  * Error return values can be the result of the mailbox command, -EINVAL
1286  * when security requirements are not met or invalid contexts, or -EBUSY
1287  * if the sanitize operation is already in flight.
1288  *
1289  * See CXL 3.0 @8.2.9.8.5.1 Sanitize and @8.2.9.8.5.2 Secure Erase.
1290  */
cxl_mem_sanitize(struct cxl_memdev * cxlmd,u16 cmd)1291 int cxl_mem_sanitize(struct cxl_memdev *cxlmd, u16 cmd)
1292 {
1293 	struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlmd->cxlds);
1294 	struct cxl_port  *endpoint;
1295 
1296 	/* synchronize with cxl_mem_probe() and decoder write operations */
1297 	guard(device)(&cxlmd->dev);
1298 	endpoint = cxlmd->endpoint;
1299 	guard(rwsem_read)(&cxl_rwsem.region);
1300 	/*
1301 	 * Require an endpoint to be safe otherwise the driver can not
1302 	 * be sure that the device is unmapped.
1303 	 */
1304 	if (endpoint && cxl_num_decoders_committed(endpoint) == 0)
1305 		return __cxl_mem_sanitize(mds, cmd);
1306 
1307 	return -EBUSY;
1308 }
1309 
add_part(struct cxl_dpa_info * info,u64 start,u64 size,enum cxl_partition_mode mode)1310 static void add_part(struct cxl_dpa_info *info, u64 start, u64 size, enum cxl_partition_mode mode)
1311 {
1312 	int i = info->nr_partitions;
1313 
1314 	if (size == 0)
1315 		return;
1316 
1317 	info->part[i].range = (struct range) {
1318 		.start = start,
1319 		.end = start + size - 1,
1320 	};
1321 	info->part[i].mode = mode;
1322 	info->nr_partitions++;
1323 }
1324 
cxl_mem_dpa_fetch(struct cxl_memdev_state * mds,struct cxl_dpa_info * info)1325 int cxl_mem_dpa_fetch(struct cxl_memdev_state *mds, struct cxl_dpa_info *info)
1326 {
1327 	struct cxl_dev_state *cxlds = &mds->cxlds;
1328 	struct device *dev = cxlds->dev;
1329 	int rc;
1330 
1331 	if (!cxlds->media_ready) {
1332 		info->size = 0;
1333 		return 0;
1334 	}
1335 
1336 	info->size = mds->total_bytes;
1337 
1338 	if (mds->partition_align_bytes == 0) {
1339 		add_part(info, 0, mds->volatile_only_bytes, CXL_PARTMODE_RAM);
1340 		add_part(info, mds->volatile_only_bytes,
1341 			 mds->persistent_only_bytes, CXL_PARTMODE_PMEM);
1342 		return 0;
1343 	}
1344 
1345 	rc = cxl_mem_get_partition_info(mds);
1346 	if (rc) {
1347 		dev_err(dev, "Failed to query partition information\n");
1348 		return rc;
1349 	}
1350 
1351 	add_part(info, 0, mds->active_volatile_bytes, CXL_PARTMODE_RAM);
1352 	add_part(info, mds->active_volatile_bytes, mds->active_persistent_bytes,
1353 		 CXL_PARTMODE_PMEM);
1354 
1355 	return 0;
1356 }
1357 EXPORT_SYMBOL_NS_GPL(cxl_mem_dpa_fetch, "CXL");
1358 
cxl_get_dirty_count(struct cxl_memdev_state * mds,u32 * count)1359 int cxl_get_dirty_count(struct cxl_memdev_state *mds, u32 *count)
1360 {
1361 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1362 	struct cxl_mbox_get_health_info_out hi;
1363 	struct cxl_mbox_cmd mbox_cmd;
1364 	int rc;
1365 
1366 	mbox_cmd = (struct cxl_mbox_cmd) {
1367 		.opcode = CXL_MBOX_OP_GET_HEALTH_INFO,
1368 		.size_out = sizeof(hi),
1369 		.payload_out = &hi,
1370 	};
1371 
1372 	rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1373 	if (!rc)
1374 		*count = le32_to_cpu(hi.dirty_shutdown_cnt);
1375 
1376 	return rc;
1377 }
1378 EXPORT_SYMBOL_NS_GPL(cxl_get_dirty_count, "CXL");
1379 
cxl_arm_dirty_shutdown(struct cxl_memdev_state * mds)1380 int cxl_arm_dirty_shutdown(struct cxl_memdev_state *mds)
1381 {
1382 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1383 	struct cxl_mbox_cmd mbox_cmd;
1384 	struct cxl_mbox_set_shutdown_state_in in = {
1385 		.state = 1
1386 	};
1387 
1388 	mbox_cmd = (struct cxl_mbox_cmd) {
1389 		.opcode = CXL_MBOX_OP_SET_SHUTDOWN_STATE,
1390 		.size_in = sizeof(in),
1391 		.payload_in = &in,
1392 	};
1393 
1394 	return cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1395 }
1396 EXPORT_SYMBOL_NS_GPL(cxl_arm_dirty_shutdown, "CXL");
1397 
cxl_set_timestamp(struct cxl_memdev_state * mds)1398 int cxl_set_timestamp(struct cxl_memdev_state *mds)
1399 {
1400 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1401 	struct cxl_mbox_cmd mbox_cmd;
1402 	struct cxl_mbox_set_timestamp_in pi;
1403 	int rc;
1404 
1405 	pi.timestamp = cpu_to_le64(ktime_get_real_ns());
1406 	mbox_cmd = (struct cxl_mbox_cmd) {
1407 		.opcode = CXL_MBOX_OP_SET_TIMESTAMP,
1408 		.size_in = sizeof(pi),
1409 		.payload_in = &pi,
1410 	};
1411 
1412 	rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1413 	/*
1414 	 * Command is optional. Devices may have another way of providing
1415 	 * a timestamp, or may return all 0s in timestamp fields.
1416 	 * Don't report an error if this command isn't supported
1417 	 */
1418 	if (rc && (mbox_cmd.return_code != CXL_MBOX_CMD_RC_UNSUPPORTED))
1419 		return rc;
1420 
1421 	return 0;
1422 }
1423 EXPORT_SYMBOL_NS_GPL(cxl_set_timestamp, "CXL");
1424 
cxl_mem_get_poison(struct cxl_memdev * cxlmd,u64 offset,u64 len,struct cxl_region * cxlr)1425 int cxl_mem_get_poison(struct cxl_memdev *cxlmd, u64 offset, u64 len,
1426 		       struct cxl_region *cxlr)
1427 {
1428 	struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlmd->cxlds);
1429 	struct cxl_mailbox *cxl_mbox = &cxlmd->cxlds->cxl_mbox;
1430 	struct cxl_mbox_poison_out *po;
1431 	struct cxl_mbox_poison_in pi;
1432 	int nr_records = 0;
1433 	int rc;
1434 
1435 	ACQUIRE(mutex_intr, lock)(&mds->poison.mutex);
1436 	if ((rc = ACQUIRE_ERR(mutex_intr, &lock)))
1437 		return rc;
1438 
1439 	po = mds->poison.list_out;
1440 	pi.offset = cpu_to_le64(offset);
1441 	pi.length = cpu_to_le64(len / CXL_POISON_LEN_MULT);
1442 
1443 	do {
1444 		struct cxl_mbox_cmd mbox_cmd = (struct cxl_mbox_cmd){
1445 			.opcode = CXL_MBOX_OP_GET_POISON,
1446 			.size_in = sizeof(pi),
1447 			.payload_in = &pi,
1448 			.size_out = cxl_mbox->payload_size,
1449 			.payload_out = po,
1450 			.min_out = struct_size(po, record, 0),
1451 		};
1452 
1453 		rc = cxl_internal_send_cmd(cxl_mbox, &mbox_cmd);
1454 		if (rc)
1455 			break;
1456 
1457 		for (int i = 0; i < le16_to_cpu(po->count); i++)
1458 			trace_cxl_poison(cxlmd, cxlr, &po->record[i],
1459 					 po->flags, po->overflow_ts,
1460 					 CXL_POISON_TRACE_LIST);
1461 
1462 		/* Protect against an uncleared _FLAG_MORE */
1463 		nr_records = nr_records + le16_to_cpu(po->count);
1464 		if (nr_records >= mds->poison.max_errors) {
1465 			dev_dbg(&cxlmd->dev, "Max Error Records reached: %d\n",
1466 				nr_records);
1467 			break;
1468 		}
1469 	} while (po->flags & CXL_POISON_FLAG_MORE);
1470 
1471 	return rc;
1472 }
1473 EXPORT_SYMBOL_NS_GPL(cxl_mem_get_poison, "CXL");
1474 
free_poison_buf(void * buf)1475 static void free_poison_buf(void *buf)
1476 {
1477 	kvfree(buf);
1478 }
1479 
1480 /* Get Poison List output buffer is protected by mds->poison.lock */
cxl_poison_alloc_buf(struct cxl_memdev_state * mds)1481 static int cxl_poison_alloc_buf(struct cxl_memdev_state *mds)
1482 {
1483 	struct cxl_mailbox *cxl_mbox = &mds->cxlds.cxl_mbox;
1484 
1485 	mds->poison.list_out = kvmalloc(cxl_mbox->payload_size, GFP_KERNEL);
1486 	if (!mds->poison.list_out)
1487 		return -ENOMEM;
1488 
1489 	return devm_add_action_or_reset(mds->cxlds.dev, free_poison_buf,
1490 					mds->poison.list_out);
1491 }
1492 
cxl_poison_state_init(struct cxl_memdev_state * mds)1493 int cxl_poison_state_init(struct cxl_memdev_state *mds)
1494 {
1495 	int rc;
1496 
1497 	if (!test_bit(CXL_POISON_ENABLED_LIST, mds->poison.enabled_cmds))
1498 		return 0;
1499 
1500 	rc = cxl_poison_alloc_buf(mds);
1501 	if (rc) {
1502 		clear_bit(CXL_POISON_ENABLED_LIST, mds->poison.enabled_cmds);
1503 		return rc;
1504 	}
1505 
1506 	mutex_init(&mds->poison.mutex);
1507 	return 0;
1508 }
1509 EXPORT_SYMBOL_NS_GPL(cxl_poison_state_init, "CXL");
1510 
cxl_mailbox_init(struct cxl_mailbox * cxl_mbox,struct device * host)1511 int cxl_mailbox_init(struct cxl_mailbox *cxl_mbox, struct device *host)
1512 {
1513 	if (!cxl_mbox || !host)
1514 		return -EINVAL;
1515 
1516 	cxl_mbox->host = host;
1517 	mutex_init(&cxl_mbox->mbox_mutex);
1518 	rcuwait_init(&cxl_mbox->mbox_wait);
1519 
1520 	return 0;
1521 }
1522 EXPORT_SYMBOL_NS_GPL(cxl_mailbox_init, "CXL");
1523 
cxl_memdev_state_create(struct device * dev)1524 struct cxl_memdev_state *cxl_memdev_state_create(struct device *dev)
1525 {
1526 	struct cxl_memdev_state *mds;
1527 	int rc;
1528 
1529 	mds = devm_kzalloc(dev, sizeof(*mds), GFP_KERNEL);
1530 	if (!mds) {
1531 		dev_err(dev, "No memory available\n");
1532 		return ERR_PTR(-ENOMEM);
1533 	}
1534 
1535 	mutex_init(&mds->event.log_lock);
1536 	mds->cxlds.dev = dev;
1537 	mds->cxlds.reg_map.host = dev;
1538 	mds->cxlds.cxl_mbox.host = dev;
1539 	mds->cxlds.reg_map.resource = CXL_RESOURCE_NONE;
1540 	mds->cxlds.type = CXL_DEVTYPE_CLASSMEM;
1541 
1542 	rc = devm_cxl_register_mce_notifier(dev, &mds->mce_notifier);
1543 	if (rc == -EOPNOTSUPP)
1544 		dev_warn(dev, "CXL MCE unsupported\n");
1545 	else if (rc)
1546 		return ERR_PTR(rc);
1547 
1548 	return mds;
1549 }
1550 EXPORT_SYMBOL_NS_GPL(cxl_memdev_state_create, "CXL");
1551 
cxl_mbox_init(void)1552 void __init cxl_mbox_init(void)
1553 {
1554 	struct dentry *mbox_debugfs;
1555 
1556 	mbox_debugfs = cxl_debugfs_create_dir("mbox");
1557 	debugfs_create_bool("raw_allow_all", 0600, mbox_debugfs,
1558 			    &cxl_raw_allow_all);
1559 }
1560