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