Linux Audio

Check our new training course

Loading...
Note: File does not exist in v5.9.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/* Copyright(c) 2020 Intel Corporation. All rights reserved. */
   3#include <linux/platform_device.h>
   4#include <linux/memregion.h>
   5#include <linux/workqueue.h>
   6#include <linux/debugfs.h>
   7#include <linux/device.h>
   8#include <linux/module.h>
   9#include <linux/pci.h>
  10#include <linux/slab.h>
  11#include <linux/idr.h>
  12#include <linux/node.h>
  13#include <cxlmem.h>
  14#include <cxlpci.h>
  15#include <cxl.h>
  16#include "core.h"
  17
  18/**
  19 * DOC: cxl core
  20 *
  21 * The CXL core provides a set of interfaces that can be consumed by CXL aware
  22 * drivers. The interfaces allow for creation, modification, and destruction of
  23 * regions, memory devices, ports, and decoders. CXL aware drivers must register
  24 * with the CXL core via these interfaces in order to be able to participate in
  25 * cross-device interleave coordination. The CXL core also establishes and
  26 * maintains the bridge to the nvdimm subsystem.
  27 *
  28 * CXL core introduces sysfs hierarchy to control the devices that are
  29 * instantiated by the core.
  30 */
  31
  32/*
  33 * All changes to the interleave configuration occur with this lock held
  34 * for write.
  35 */
  36DECLARE_RWSEM(cxl_region_rwsem);
  37
  38static DEFINE_IDA(cxl_port_ida);
  39static DEFINE_XARRAY(cxl_root_buses);
  40
  41int cxl_num_decoders_committed(struct cxl_port *port)
  42{
  43	lockdep_assert_held(&cxl_region_rwsem);
  44
  45	return port->commit_end + 1;
  46}
  47
  48static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
  49			    char *buf)
  50{
  51	return sysfs_emit(buf, "%s\n", dev->type->name);
  52}
  53static DEVICE_ATTR_RO(devtype);
  54
  55static int cxl_device_id(const struct device *dev)
  56{
  57	if (dev->type == &cxl_nvdimm_bridge_type)
  58		return CXL_DEVICE_NVDIMM_BRIDGE;
  59	if (dev->type == &cxl_nvdimm_type)
  60		return CXL_DEVICE_NVDIMM;
  61	if (dev->type == CXL_PMEM_REGION_TYPE())
  62		return CXL_DEVICE_PMEM_REGION;
  63	if (dev->type == CXL_DAX_REGION_TYPE())
  64		return CXL_DEVICE_DAX_REGION;
  65	if (is_cxl_port(dev)) {
  66		if (is_cxl_root(to_cxl_port(dev)))
  67			return CXL_DEVICE_ROOT;
  68		return CXL_DEVICE_PORT;
  69	}
  70	if (is_cxl_memdev(dev))
  71		return CXL_DEVICE_MEMORY_EXPANDER;
  72	if (dev->type == CXL_REGION_TYPE())
  73		return CXL_DEVICE_REGION;
  74	if (dev->type == &cxl_pmu_type)
  75		return CXL_DEVICE_PMU;
  76	return 0;
  77}
  78
  79static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
  80			     char *buf)
  81{
  82	return sysfs_emit(buf, CXL_MODALIAS_FMT "\n", cxl_device_id(dev));
  83}
  84static DEVICE_ATTR_RO(modalias);
  85
  86static struct attribute *cxl_base_attributes[] = {
  87	&dev_attr_devtype.attr,
  88	&dev_attr_modalias.attr,
  89	NULL,
  90};
  91
  92struct attribute_group cxl_base_attribute_group = {
  93	.attrs = cxl_base_attributes,
  94};
  95
  96static ssize_t start_show(struct device *dev, struct device_attribute *attr,
  97			  char *buf)
  98{
  99	struct cxl_decoder *cxld = to_cxl_decoder(dev);
 100
 101	return sysfs_emit(buf, "%#llx\n", cxld->hpa_range.start);
 102}
 103static DEVICE_ATTR_ADMIN_RO(start);
 104
 105static ssize_t size_show(struct device *dev, struct device_attribute *attr,
 106			char *buf)
 107{
 108	struct cxl_decoder *cxld = to_cxl_decoder(dev);
 109
 110	return sysfs_emit(buf, "%#llx\n", range_len(&cxld->hpa_range));
 111}
 112static DEVICE_ATTR_RO(size);
 113
 114#define CXL_DECODER_FLAG_ATTR(name, flag)                            \
 115static ssize_t name##_show(struct device *dev,                       \
 116			   struct device_attribute *attr, char *buf) \
 117{                                                                    \
 118	struct cxl_decoder *cxld = to_cxl_decoder(dev);              \
 119                                                                     \
 120	return sysfs_emit(buf, "%s\n",                               \
 121			  (cxld->flags & (flag)) ? "1" : "0");       \
 122}                                                                    \
 123static DEVICE_ATTR_RO(name)
 124
 125CXL_DECODER_FLAG_ATTR(cap_pmem, CXL_DECODER_F_PMEM);
 126CXL_DECODER_FLAG_ATTR(cap_ram, CXL_DECODER_F_RAM);
 127CXL_DECODER_FLAG_ATTR(cap_type2, CXL_DECODER_F_TYPE2);
 128CXL_DECODER_FLAG_ATTR(cap_type3, CXL_DECODER_F_TYPE3);
 129CXL_DECODER_FLAG_ATTR(locked, CXL_DECODER_F_LOCK);
 130
 131static ssize_t target_type_show(struct device *dev,
 132				struct device_attribute *attr, char *buf)
 133{
 134	struct cxl_decoder *cxld = to_cxl_decoder(dev);
 135
 136	switch (cxld->target_type) {
 137	case CXL_DECODER_DEVMEM:
 138		return sysfs_emit(buf, "accelerator\n");
 139	case CXL_DECODER_HOSTONLYMEM:
 140		return sysfs_emit(buf, "expander\n");
 141	}
 142	return -ENXIO;
 143}
 144static DEVICE_ATTR_RO(target_type);
 145
 146static ssize_t emit_target_list(struct cxl_switch_decoder *cxlsd, char *buf)
 147{
 148	struct cxl_decoder *cxld = &cxlsd->cxld;
 149	ssize_t offset = 0;
 150	int i, rc = 0;
 151
 152	for (i = 0; i < cxld->interleave_ways; i++) {
 153		struct cxl_dport *dport = cxlsd->target[i];
 154		struct cxl_dport *next = NULL;
 155
 156		if (!dport)
 157			break;
 158
 159		if (i + 1 < cxld->interleave_ways)
 160			next = cxlsd->target[i + 1];
 161		rc = sysfs_emit_at(buf, offset, "%d%s", dport->port_id,
 162				   next ? "," : "");
 163		if (rc < 0)
 164			return rc;
 165		offset += rc;
 166	}
 167
 168	return offset;
 169}
 170
 171static ssize_t target_list_show(struct device *dev,
 172				struct device_attribute *attr, char *buf)
 173{
 174	struct cxl_switch_decoder *cxlsd = to_cxl_switch_decoder(dev);
 175	ssize_t offset;
 176	int rc;
 177
 178	guard(rwsem_read)(&cxl_region_rwsem);
 179	rc = emit_target_list(cxlsd, buf);
 180	if (rc < 0)
 181		return rc;
 182	offset = rc;
 183
 184	rc = sysfs_emit_at(buf, offset, "\n");
 185	if (rc < 0)
 186		return rc;
 187
 188	return offset + rc;
 189}
 190static DEVICE_ATTR_RO(target_list);
 191
 192static ssize_t mode_show(struct device *dev, struct device_attribute *attr,
 193			 char *buf)
 194{
 195	struct cxl_endpoint_decoder *cxled = to_cxl_endpoint_decoder(dev);
 196
 197	return sysfs_emit(buf, "%s\n", cxl_decoder_mode_name(cxled->mode));
 198}
 199
 200static ssize_t mode_store(struct device *dev, struct device_attribute *attr,
 201			  const char *buf, size_t len)
 202{
 203	struct cxl_endpoint_decoder *cxled = to_cxl_endpoint_decoder(dev);
 204	enum cxl_decoder_mode mode;
 205	ssize_t rc;
 206
 207	if (sysfs_streq(buf, "pmem"))
 208		mode = CXL_DECODER_PMEM;
 209	else if (sysfs_streq(buf, "ram"))
 210		mode = CXL_DECODER_RAM;
 211	else
 212		return -EINVAL;
 213
 214	rc = cxl_dpa_set_mode(cxled, mode);
 215	if (rc)
 216		return rc;
 217
 218	return len;
 219}
 220static DEVICE_ATTR_RW(mode);
 221
 222static ssize_t dpa_resource_show(struct device *dev, struct device_attribute *attr,
 223			    char *buf)
 224{
 225	struct cxl_endpoint_decoder *cxled = to_cxl_endpoint_decoder(dev);
 226
 227	guard(rwsem_read)(&cxl_dpa_rwsem);
 228	return sysfs_emit(buf, "%#llx\n", (u64)cxl_dpa_resource_start(cxled));
 229}
 230static DEVICE_ATTR_RO(dpa_resource);
 231
 232static ssize_t dpa_size_show(struct device *dev, struct device_attribute *attr,
 233			     char *buf)
 234{
 235	struct cxl_endpoint_decoder *cxled = to_cxl_endpoint_decoder(dev);
 236	resource_size_t size = cxl_dpa_size(cxled);
 237
 238	return sysfs_emit(buf, "%pa\n", &size);
 239}
 240
 241static ssize_t dpa_size_store(struct device *dev, struct device_attribute *attr,
 242			      const char *buf, size_t len)
 243{
 244	struct cxl_endpoint_decoder *cxled = to_cxl_endpoint_decoder(dev);
 245	unsigned long long size;
 246	ssize_t rc;
 247
 248	rc = kstrtoull(buf, 0, &size);
 249	if (rc)
 250		return rc;
 251
 252	if (!IS_ALIGNED(size, SZ_256M))
 253		return -EINVAL;
 254
 255	rc = cxl_dpa_free(cxled);
 256	if (rc)
 257		return rc;
 258
 259	if (size == 0)
 260		return len;
 261
 262	rc = cxl_dpa_alloc(cxled, size);
 263	if (rc)
 264		return rc;
 265
 266	return len;
 267}
 268static DEVICE_ATTR_RW(dpa_size);
 269
 270static ssize_t interleave_granularity_show(struct device *dev,
 271					   struct device_attribute *attr,
 272					   char *buf)
 273{
 274	struct cxl_decoder *cxld = to_cxl_decoder(dev);
 275
 276	return sysfs_emit(buf, "%d\n", cxld->interleave_granularity);
 277}
 278
 279static DEVICE_ATTR_RO(interleave_granularity);
 280
 281static ssize_t interleave_ways_show(struct device *dev,
 282				    struct device_attribute *attr, char *buf)
 283{
 284	struct cxl_decoder *cxld = to_cxl_decoder(dev);
 285
 286	return sysfs_emit(buf, "%d\n", cxld->interleave_ways);
 287}
 288
 289static DEVICE_ATTR_RO(interleave_ways);
 290
 291static ssize_t qos_class_show(struct device *dev,
 292			      struct device_attribute *attr, char *buf)
 293{
 294	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(dev);
 295
 296	return sysfs_emit(buf, "%d\n", cxlrd->qos_class);
 297}
 298static DEVICE_ATTR_RO(qos_class);
 299
 300static struct attribute *cxl_decoder_base_attrs[] = {
 301	&dev_attr_start.attr,
 302	&dev_attr_size.attr,
 303	&dev_attr_locked.attr,
 304	&dev_attr_interleave_granularity.attr,
 305	&dev_attr_interleave_ways.attr,
 306	NULL,
 307};
 308
 309static struct attribute_group cxl_decoder_base_attribute_group = {
 310	.attrs = cxl_decoder_base_attrs,
 311};
 312
 313static struct attribute *cxl_decoder_root_attrs[] = {
 314	&dev_attr_cap_pmem.attr,
 315	&dev_attr_cap_ram.attr,
 316	&dev_attr_cap_type2.attr,
 317	&dev_attr_cap_type3.attr,
 318	&dev_attr_target_list.attr,
 319	&dev_attr_qos_class.attr,
 320	SET_CXL_REGION_ATTR(create_pmem_region)
 321	SET_CXL_REGION_ATTR(create_ram_region)
 322	SET_CXL_REGION_ATTR(delete_region)
 323	NULL,
 324};
 325
 326static bool can_create_pmem(struct cxl_root_decoder *cxlrd)
 327{
 328	unsigned long flags = CXL_DECODER_F_TYPE3 | CXL_DECODER_F_PMEM;
 329
 330	return (cxlrd->cxlsd.cxld.flags & flags) == flags;
 331}
 332
 333static bool can_create_ram(struct cxl_root_decoder *cxlrd)
 334{
 335	unsigned long flags = CXL_DECODER_F_TYPE3 | CXL_DECODER_F_RAM;
 336
 337	return (cxlrd->cxlsd.cxld.flags & flags) == flags;
 338}
 339
 340static umode_t cxl_root_decoder_visible(struct kobject *kobj, struct attribute *a, int n)
 341{
 342	struct device *dev = kobj_to_dev(kobj);
 343	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(dev);
 344
 345	if (a == CXL_REGION_ATTR(create_pmem_region) && !can_create_pmem(cxlrd))
 346		return 0;
 347
 348	if (a == CXL_REGION_ATTR(create_ram_region) && !can_create_ram(cxlrd))
 349		return 0;
 350
 351	if (a == CXL_REGION_ATTR(delete_region) &&
 352	    !(can_create_pmem(cxlrd) || can_create_ram(cxlrd)))
 353		return 0;
 354
 355	return a->mode;
 356}
 357
 358static struct attribute_group cxl_decoder_root_attribute_group = {
 359	.attrs = cxl_decoder_root_attrs,
 360	.is_visible = cxl_root_decoder_visible,
 361};
 362
 363static const struct attribute_group *cxl_decoder_root_attribute_groups[] = {
 364	&cxl_decoder_root_attribute_group,
 365	&cxl_decoder_base_attribute_group,
 366	&cxl_base_attribute_group,
 367	NULL,
 368};
 369
 370static struct attribute *cxl_decoder_switch_attrs[] = {
 371	&dev_attr_target_type.attr,
 372	&dev_attr_target_list.attr,
 373	SET_CXL_REGION_ATTR(region)
 374	NULL,
 375};
 376
 377static struct attribute_group cxl_decoder_switch_attribute_group = {
 378	.attrs = cxl_decoder_switch_attrs,
 379};
 380
 381static const struct attribute_group *cxl_decoder_switch_attribute_groups[] = {
 382	&cxl_decoder_switch_attribute_group,
 383	&cxl_decoder_base_attribute_group,
 384	&cxl_base_attribute_group,
 385	NULL,
 386};
 387
 388static struct attribute *cxl_decoder_endpoint_attrs[] = {
 389	&dev_attr_target_type.attr,
 390	&dev_attr_mode.attr,
 391	&dev_attr_dpa_size.attr,
 392	&dev_attr_dpa_resource.attr,
 393	SET_CXL_REGION_ATTR(region)
 394	NULL,
 395};
 396
 397static struct attribute_group cxl_decoder_endpoint_attribute_group = {
 398	.attrs = cxl_decoder_endpoint_attrs,
 399};
 400
 401static const struct attribute_group *cxl_decoder_endpoint_attribute_groups[] = {
 402	&cxl_decoder_base_attribute_group,
 403	&cxl_decoder_endpoint_attribute_group,
 404	&cxl_base_attribute_group,
 405	NULL,
 406};
 407
 408static void __cxl_decoder_release(struct cxl_decoder *cxld)
 409{
 410	struct cxl_port *port = to_cxl_port(cxld->dev.parent);
 411
 412	ida_free(&port->decoder_ida, cxld->id);
 413	put_device(&port->dev);
 414}
 415
 416static void cxl_endpoint_decoder_release(struct device *dev)
 417{
 418	struct cxl_endpoint_decoder *cxled = to_cxl_endpoint_decoder(dev);
 419
 420	__cxl_decoder_release(&cxled->cxld);
 421	kfree(cxled);
 422}
 423
 424static void cxl_switch_decoder_release(struct device *dev)
 425{
 426	struct cxl_switch_decoder *cxlsd = to_cxl_switch_decoder(dev);
 427
 428	__cxl_decoder_release(&cxlsd->cxld);
 429	kfree(cxlsd);
 430}
 431
 432struct cxl_root_decoder *to_cxl_root_decoder(struct device *dev)
 433{
 434	if (dev_WARN_ONCE(dev, !is_root_decoder(dev),
 435			  "not a cxl_root_decoder device\n"))
 436		return NULL;
 437	return container_of(dev, struct cxl_root_decoder, cxlsd.cxld.dev);
 438}
 439EXPORT_SYMBOL_NS_GPL(to_cxl_root_decoder, CXL);
 440
 441static void cxl_root_decoder_release(struct device *dev)
 442{
 443	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(dev);
 444
 445	if (atomic_read(&cxlrd->region_id) >= 0)
 446		memregion_free(atomic_read(&cxlrd->region_id));
 447	__cxl_decoder_release(&cxlrd->cxlsd.cxld);
 448	kfree(cxlrd);
 449}
 450
 451static const struct device_type cxl_decoder_endpoint_type = {
 452	.name = "cxl_decoder_endpoint",
 453	.release = cxl_endpoint_decoder_release,
 454	.groups = cxl_decoder_endpoint_attribute_groups,
 455};
 456
 457static const struct device_type cxl_decoder_switch_type = {
 458	.name = "cxl_decoder_switch",
 459	.release = cxl_switch_decoder_release,
 460	.groups = cxl_decoder_switch_attribute_groups,
 461};
 462
 463static const struct device_type cxl_decoder_root_type = {
 464	.name = "cxl_decoder_root",
 465	.release = cxl_root_decoder_release,
 466	.groups = cxl_decoder_root_attribute_groups,
 467};
 468
 469bool is_endpoint_decoder(struct device *dev)
 470{
 471	return dev->type == &cxl_decoder_endpoint_type;
 472}
 473EXPORT_SYMBOL_NS_GPL(is_endpoint_decoder, CXL);
 474
 475bool is_root_decoder(struct device *dev)
 476{
 477	return dev->type == &cxl_decoder_root_type;
 478}
 479EXPORT_SYMBOL_NS_GPL(is_root_decoder, CXL);
 480
 481bool is_switch_decoder(struct device *dev)
 482{
 483	return is_root_decoder(dev) || dev->type == &cxl_decoder_switch_type;
 484}
 485EXPORT_SYMBOL_NS_GPL(is_switch_decoder, CXL);
 486
 487struct cxl_decoder *to_cxl_decoder(struct device *dev)
 488{
 489	if (dev_WARN_ONCE(dev,
 490			  !is_switch_decoder(dev) && !is_endpoint_decoder(dev),
 491			  "not a cxl_decoder device\n"))
 492		return NULL;
 493	return container_of(dev, struct cxl_decoder, dev);
 494}
 495EXPORT_SYMBOL_NS_GPL(to_cxl_decoder, CXL);
 496
 497struct cxl_endpoint_decoder *to_cxl_endpoint_decoder(struct device *dev)
 498{
 499	if (dev_WARN_ONCE(dev, !is_endpoint_decoder(dev),
 500			  "not a cxl_endpoint_decoder device\n"))
 501		return NULL;
 502	return container_of(dev, struct cxl_endpoint_decoder, cxld.dev);
 503}
 504EXPORT_SYMBOL_NS_GPL(to_cxl_endpoint_decoder, CXL);
 505
 506struct cxl_switch_decoder *to_cxl_switch_decoder(struct device *dev)
 507{
 508	if (dev_WARN_ONCE(dev, !is_switch_decoder(dev),
 509			  "not a cxl_switch_decoder device\n"))
 510		return NULL;
 511	return container_of(dev, struct cxl_switch_decoder, cxld.dev);
 512}
 513EXPORT_SYMBOL_NS_GPL(to_cxl_switch_decoder, CXL);
 514
 515static void cxl_ep_release(struct cxl_ep *ep)
 516{
 517	put_device(ep->ep);
 518	kfree(ep);
 519}
 520
 521static void cxl_ep_remove(struct cxl_port *port, struct cxl_ep *ep)
 522{
 523	if (!ep)
 524		return;
 525	xa_erase(&port->endpoints, (unsigned long) ep->ep);
 526	cxl_ep_release(ep);
 527}
 528
 529static void cxl_port_release(struct device *dev)
 530{
 531	struct cxl_port *port = to_cxl_port(dev);
 532	unsigned long index;
 533	struct cxl_ep *ep;
 534
 535	xa_for_each(&port->endpoints, index, ep)
 536		cxl_ep_remove(port, ep);
 537	xa_destroy(&port->endpoints);
 538	xa_destroy(&port->dports);
 539	xa_destroy(&port->regions);
 540	ida_free(&cxl_port_ida, port->id);
 541	if (is_cxl_root(port))
 542		kfree(to_cxl_root(port));
 543	else
 544		kfree(port);
 545}
 546
 547static ssize_t decoders_committed_show(struct device *dev,
 548				       struct device_attribute *attr, char *buf)
 549{
 550	struct cxl_port *port = to_cxl_port(dev);
 551	int rc;
 552
 553	down_read(&cxl_region_rwsem);
 554	rc = sysfs_emit(buf, "%d\n", cxl_num_decoders_committed(port));
 555	up_read(&cxl_region_rwsem);
 556
 557	return rc;
 558}
 559
 560static DEVICE_ATTR_RO(decoders_committed);
 561
 562static struct attribute *cxl_port_attrs[] = {
 563	&dev_attr_decoders_committed.attr,
 564	NULL,
 565};
 566
 567static struct attribute_group cxl_port_attribute_group = {
 568	.attrs = cxl_port_attrs,
 569};
 570
 571static const struct attribute_group *cxl_port_attribute_groups[] = {
 572	&cxl_base_attribute_group,
 573	&cxl_port_attribute_group,
 574	NULL,
 575};
 576
 577static const struct device_type cxl_port_type = {
 578	.name = "cxl_port",
 579	.release = cxl_port_release,
 580	.groups = cxl_port_attribute_groups,
 581};
 582
 583bool is_cxl_port(const struct device *dev)
 584{
 585	return dev->type == &cxl_port_type;
 586}
 587EXPORT_SYMBOL_NS_GPL(is_cxl_port, CXL);
 588
 589struct cxl_port *to_cxl_port(const struct device *dev)
 590{
 591	if (dev_WARN_ONCE(dev, dev->type != &cxl_port_type,
 592			  "not a cxl_port device\n"))
 593		return NULL;
 594	return container_of(dev, struct cxl_port, dev);
 595}
 596EXPORT_SYMBOL_NS_GPL(to_cxl_port, CXL);
 597
 598static void unregister_port(void *_port)
 599{
 600	struct cxl_port *port = _port;
 601	struct cxl_port *parent;
 602	struct device *lock_dev;
 603
 604	if (is_cxl_root(port))
 605		parent = NULL;
 606	else
 607		parent = to_cxl_port(port->dev.parent);
 608
 609	/*
 610	 * CXL root port's and the first level of ports are unregistered
 611	 * under the platform firmware device lock, all other ports are
 612	 * unregistered while holding their parent port lock.
 613	 */
 614	if (!parent)
 615		lock_dev = port->uport_dev;
 616	else if (is_cxl_root(parent))
 617		lock_dev = parent->uport_dev;
 618	else
 619		lock_dev = &parent->dev;
 620
 621	device_lock_assert(lock_dev);
 622	port->dead = true;
 623	device_unregister(&port->dev);
 624}
 625
 626static void cxl_unlink_uport(void *_port)
 627{
 628	struct cxl_port *port = _port;
 629
 630	sysfs_remove_link(&port->dev.kobj, "uport");
 631}
 632
 633static int devm_cxl_link_uport(struct device *host, struct cxl_port *port)
 634{
 635	int rc;
 636
 637	rc = sysfs_create_link(&port->dev.kobj, &port->uport_dev->kobj,
 638			       "uport");
 639	if (rc)
 640		return rc;
 641	return devm_add_action_or_reset(host, cxl_unlink_uport, port);
 642}
 643
 644static void cxl_unlink_parent_dport(void *_port)
 645{
 646	struct cxl_port *port = _port;
 647
 648	sysfs_remove_link(&port->dev.kobj, "parent_dport");
 649}
 650
 651static int devm_cxl_link_parent_dport(struct device *host,
 652				      struct cxl_port *port,
 653				      struct cxl_dport *parent_dport)
 654{
 655	int rc;
 656
 657	if (!parent_dport)
 658		return 0;
 659
 660	rc = sysfs_create_link(&port->dev.kobj, &parent_dport->dport_dev->kobj,
 661			       "parent_dport");
 662	if (rc)
 663		return rc;
 664	return devm_add_action_or_reset(host, cxl_unlink_parent_dport, port);
 665}
 666
 667static struct lock_class_key cxl_port_key;
 668
 669static struct cxl_port *cxl_port_alloc(struct device *uport_dev,
 670				       struct cxl_dport *parent_dport)
 671{
 672	struct cxl_root *cxl_root __free(kfree) = NULL;
 673	struct cxl_port *port, *_port __free(kfree) = NULL;
 674	struct device *dev;
 675	int rc;
 676
 677	/* No parent_dport, root cxl_port */
 678	if (!parent_dport) {
 679		cxl_root = kzalloc(sizeof(*cxl_root), GFP_KERNEL);
 680		if (!cxl_root)
 681			return ERR_PTR(-ENOMEM);
 682	} else {
 683		_port = kzalloc(sizeof(*port), GFP_KERNEL);
 684		if (!_port)
 685			return ERR_PTR(-ENOMEM);
 686	}
 687
 688	rc = ida_alloc(&cxl_port_ida, GFP_KERNEL);
 689	if (rc < 0)
 690		return ERR_PTR(rc);
 691
 692	if (cxl_root)
 693		port = &no_free_ptr(cxl_root)->port;
 694	else
 695		port = no_free_ptr(_port);
 696
 697	port->id = rc;
 698	port->uport_dev = uport_dev;
 699
 700	/*
 701	 * The top-level cxl_port "cxl_root" does not have a cxl_port as
 702	 * its parent and it does not have any corresponding component
 703	 * registers as its decode is described by a fixed platform
 704	 * description.
 705	 */
 706	dev = &port->dev;
 707	if (parent_dport) {
 708		struct cxl_port *parent_port = parent_dport->port;
 709		struct cxl_port *iter;
 710
 711		dev->parent = &parent_port->dev;
 712		port->depth = parent_port->depth + 1;
 713		port->parent_dport = parent_dport;
 714
 715		/*
 716		 * walk to the host bridge, or the first ancestor that knows
 717		 * the host bridge
 718		 */
 719		iter = port;
 720		while (!iter->host_bridge &&
 721		       !is_cxl_root(to_cxl_port(iter->dev.parent)))
 722			iter = to_cxl_port(iter->dev.parent);
 723		if (iter->host_bridge)
 724			port->host_bridge = iter->host_bridge;
 725		else if (parent_dport->rch)
 726			port->host_bridge = parent_dport->dport_dev;
 727		else
 728			port->host_bridge = iter->uport_dev;
 729		dev_dbg(uport_dev, "host-bridge: %s\n",
 730			dev_name(port->host_bridge));
 731	} else
 732		dev->parent = uport_dev;
 733
 734	ida_init(&port->decoder_ida);
 735	port->hdm_end = -1;
 736	port->commit_end = -1;
 737	xa_init(&port->dports);
 738	xa_init(&port->endpoints);
 739	xa_init(&port->regions);
 740
 741	device_initialize(dev);
 742	lockdep_set_class_and_subclass(&dev->mutex, &cxl_port_key, port->depth);
 743	device_set_pm_not_required(dev);
 744	dev->bus = &cxl_bus_type;
 745	dev->type = &cxl_port_type;
 746
 747	return port;
 748}
 749
 750static int cxl_setup_comp_regs(struct device *host, struct cxl_register_map *map,
 751			       resource_size_t component_reg_phys)
 752{
 753	*map = (struct cxl_register_map) {
 754		.host = host,
 755		.reg_type = CXL_REGLOC_RBI_EMPTY,
 756		.resource = component_reg_phys,
 757	};
 758
 759	if (component_reg_phys == CXL_RESOURCE_NONE)
 760		return 0;
 761
 762	map->reg_type = CXL_REGLOC_RBI_COMPONENT;
 763	map->max_size = CXL_COMPONENT_REG_BLOCK_SIZE;
 764
 765	return cxl_setup_regs(map);
 766}
 767
 768static int cxl_port_setup_regs(struct cxl_port *port,
 769			resource_size_t component_reg_phys)
 770{
 771	if (dev_is_platform(port->uport_dev))
 772		return 0;
 773	return cxl_setup_comp_regs(&port->dev, &port->reg_map,
 774				   component_reg_phys);
 775}
 776
 777static int cxl_dport_setup_regs(struct device *host, struct cxl_dport *dport,
 778				resource_size_t component_reg_phys)
 779{
 780	int rc;
 781
 782	if (dev_is_platform(dport->dport_dev))
 783		return 0;
 784
 785	/*
 786	 * use @dport->dport_dev for the context for error messages during
 787	 * register probing, and fixup @host after the fact, since @host may be
 788	 * NULL.
 789	 */
 790	rc = cxl_setup_comp_regs(dport->dport_dev, &dport->reg_map,
 791				 component_reg_phys);
 792	dport->reg_map.host = host;
 793	return rc;
 794}
 795
 796static struct cxl_port *__devm_cxl_add_port(struct device *host,
 797					    struct device *uport_dev,
 798					    resource_size_t component_reg_phys,
 799					    struct cxl_dport *parent_dport)
 800{
 801	struct cxl_port *port;
 802	struct device *dev;
 803	int rc;
 804
 805	port = cxl_port_alloc(uport_dev, parent_dport);
 806	if (IS_ERR(port))
 807		return port;
 808
 809	dev = &port->dev;
 810	if (is_cxl_memdev(uport_dev)) {
 811		struct cxl_memdev *cxlmd = to_cxl_memdev(uport_dev);
 812		struct cxl_dev_state *cxlds = cxlmd->cxlds;
 813
 814		rc = dev_set_name(dev, "endpoint%d", port->id);
 815		if (rc)
 816			goto err;
 817
 818		/*
 819		 * The endpoint driver already enumerated the component and RAS
 820		 * registers. Reuse that enumeration while prepping them to be
 821		 * mapped by the cxl_port driver.
 822		 */
 823		port->reg_map = cxlds->reg_map;
 824		port->reg_map.host = &port->dev;
 825	} else if (parent_dport) {
 826		rc = dev_set_name(dev, "port%d", port->id);
 827		if (rc)
 828			goto err;
 829
 830		rc = cxl_port_setup_regs(port, component_reg_phys);
 831		if (rc)
 832			goto err;
 833	} else
 834		rc = dev_set_name(dev, "root%d", port->id);
 835	if (rc)
 836		goto err;
 837
 838	rc = device_add(dev);
 839	if (rc)
 840		goto err;
 841
 842	rc = devm_add_action_or_reset(host, unregister_port, port);
 843	if (rc)
 844		return ERR_PTR(rc);
 845
 846	rc = devm_cxl_link_uport(host, port);
 847	if (rc)
 848		return ERR_PTR(rc);
 849
 850	rc = devm_cxl_link_parent_dport(host, port, parent_dport);
 851	if (rc)
 852		return ERR_PTR(rc);
 853
 854	if (parent_dport && dev_is_pci(uport_dev))
 855		port->pci_latency = cxl_pci_get_latency(to_pci_dev(uport_dev));
 856
 857	return port;
 858
 859err:
 860	put_device(dev);
 861	return ERR_PTR(rc);
 862}
 863
 864/**
 865 * devm_cxl_add_port - register a cxl_port in CXL memory decode hierarchy
 866 * @host: host device for devm operations
 867 * @uport_dev: "physical" device implementing this upstream port
 868 * @component_reg_phys: (optional) for configurable cxl_port instances
 869 * @parent_dport: next hop up in the CXL memory decode hierarchy
 870 */
 871struct cxl_port *devm_cxl_add_port(struct device *host,
 872				   struct device *uport_dev,
 873				   resource_size_t component_reg_phys,
 874				   struct cxl_dport *parent_dport)
 875{
 876	struct cxl_port *port, *parent_port;
 877
 878	port = __devm_cxl_add_port(host, uport_dev, component_reg_phys,
 879				   parent_dport);
 880
 881	parent_port = parent_dport ? parent_dport->port : NULL;
 882	if (IS_ERR(port)) {
 883		dev_dbg(uport_dev, "Failed to add%s%s%s: %ld\n",
 884			parent_port ? " port to " : "",
 885			parent_port ? dev_name(&parent_port->dev) : "",
 886			parent_port ? "" : " root port",
 887			PTR_ERR(port));
 888	} else {
 889		dev_dbg(uport_dev, "%s added%s%s%s\n",
 890			dev_name(&port->dev),
 891			parent_port ? " to " : "",
 892			parent_port ? dev_name(&parent_port->dev) : "",
 893			parent_port ? "" : " (root port)");
 894	}
 895
 896	return port;
 897}
 898EXPORT_SYMBOL_NS_GPL(devm_cxl_add_port, CXL);
 899
 900struct cxl_root *devm_cxl_add_root(struct device *host,
 901				   const struct cxl_root_ops *ops)
 902{
 903	struct cxl_root *cxl_root;
 904	struct cxl_port *port;
 905
 906	port = devm_cxl_add_port(host, host, CXL_RESOURCE_NONE, NULL);
 907	if (IS_ERR(port))
 908		return (struct cxl_root *)port;
 909
 910	cxl_root = to_cxl_root(port);
 911	cxl_root->ops = ops;
 912	return cxl_root;
 913}
 914EXPORT_SYMBOL_NS_GPL(devm_cxl_add_root, CXL);
 915
 916struct pci_bus *cxl_port_to_pci_bus(struct cxl_port *port)
 917{
 918	/* There is no pci_bus associated with a CXL platform-root port */
 919	if (is_cxl_root(port))
 920		return NULL;
 921
 922	if (dev_is_pci(port->uport_dev)) {
 923		struct pci_dev *pdev = to_pci_dev(port->uport_dev);
 924
 925		return pdev->subordinate;
 926	}
 927
 928	return xa_load(&cxl_root_buses, (unsigned long)port->uport_dev);
 929}
 930EXPORT_SYMBOL_NS_GPL(cxl_port_to_pci_bus, CXL);
 931
 932static void unregister_pci_bus(void *uport_dev)
 933{
 934	xa_erase(&cxl_root_buses, (unsigned long)uport_dev);
 935}
 936
 937int devm_cxl_register_pci_bus(struct device *host, struct device *uport_dev,
 938			      struct pci_bus *bus)
 939{
 940	int rc;
 941
 942	if (dev_is_pci(uport_dev))
 943		return -EINVAL;
 944
 945	rc = xa_insert(&cxl_root_buses, (unsigned long)uport_dev, bus,
 946		       GFP_KERNEL);
 947	if (rc)
 948		return rc;
 949	return devm_add_action_or_reset(host, unregister_pci_bus, uport_dev);
 950}
 951EXPORT_SYMBOL_NS_GPL(devm_cxl_register_pci_bus, CXL);
 952
 953static bool dev_is_cxl_root_child(struct device *dev)
 954{
 955	struct cxl_port *port, *parent;
 956
 957	if (!is_cxl_port(dev))
 958		return false;
 959
 960	port = to_cxl_port(dev);
 961	if (is_cxl_root(port))
 962		return false;
 963
 964	parent = to_cxl_port(port->dev.parent);
 965	if (is_cxl_root(parent))
 966		return true;
 967
 968	return false;
 969}
 970
 971struct cxl_root *find_cxl_root(struct cxl_port *port)
 972{
 973	struct cxl_port *iter = port;
 974
 975	while (iter && !is_cxl_root(iter))
 976		iter = to_cxl_port(iter->dev.parent);
 977
 978	if (!iter)
 979		return NULL;
 980	get_device(&iter->dev);
 981	return to_cxl_root(iter);
 982}
 983EXPORT_SYMBOL_NS_GPL(find_cxl_root, CXL);
 984
 985void put_cxl_root(struct cxl_root *cxl_root)
 986{
 987	if (!cxl_root)
 988		return;
 989
 990	put_device(&cxl_root->port.dev);
 991}
 992EXPORT_SYMBOL_NS_GPL(put_cxl_root, CXL);
 993
 994static struct cxl_dport *find_dport(struct cxl_port *port, int id)
 995{
 996	struct cxl_dport *dport;
 997	unsigned long index;
 998
 999	device_lock_assert(&port->dev);
1000	xa_for_each(&port->dports, index, dport)
1001		if (dport->port_id == id)
1002			return dport;
1003	return NULL;
1004}
1005
1006static int add_dport(struct cxl_port *port, struct cxl_dport *dport)
1007{
1008	struct cxl_dport *dup;
1009	int rc;
1010
1011	device_lock_assert(&port->dev);
1012	dup = find_dport(port, dport->port_id);
1013	if (dup) {
1014		dev_err(&port->dev,
1015			"unable to add dport%d-%s non-unique port id (%s)\n",
1016			dport->port_id, dev_name(dport->dport_dev),
1017			dev_name(dup->dport_dev));
1018		return -EBUSY;
1019	}
1020
1021	rc = xa_insert(&port->dports, (unsigned long)dport->dport_dev, dport,
1022		       GFP_KERNEL);
1023	if (rc)
1024		return rc;
1025
1026	port->nr_dports++;
1027	return 0;
1028}
1029
1030/*
1031 * Since root-level CXL dports cannot be enumerated by PCI they are not
1032 * enumerated by the common port driver that acquires the port lock over
1033 * dport add/remove. Instead, root dports are manually added by a
1034 * platform driver and cond_cxl_root_lock() is used to take the missing
1035 * port lock in that case.
1036 */
1037static void cond_cxl_root_lock(struct cxl_port *port)
1038{
1039	if (is_cxl_root(port))
1040		device_lock(&port->dev);
1041}
1042
1043static void cond_cxl_root_unlock(struct cxl_port *port)
1044{
1045	if (is_cxl_root(port))
1046		device_unlock(&port->dev);
1047}
1048
1049static void cxl_dport_remove(void *data)
1050{
1051	struct cxl_dport *dport = data;
1052	struct cxl_port *port = dport->port;
1053
1054	xa_erase(&port->dports, (unsigned long) dport->dport_dev);
1055	put_device(dport->dport_dev);
1056}
1057
1058static void cxl_dport_unlink(void *data)
1059{
1060	struct cxl_dport *dport = data;
1061	struct cxl_port *port = dport->port;
1062	char link_name[CXL_TARGET_STRLEN];
1063
1064	sprintf(link_name, "dport%d", dport->port_id);
1065	sysfs_remove_link(&port->dev.kobj, link_name);
1066}
1067
1068static struct cxl_dport *
1069__devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
1070		     int port_id, resource_size_t component_reg_phys,
1071		     resource_size_t rcrb)
1072{
1073	char link_name[CXL_TARGET_STRLEN];
1074	struct cxl_dport *dport;
1075	struct device *host;
1076	int rc;
1077
1078	if (is_cxl_root(port))
1079		host = port->uport_dev;
1080	else
1081		host = &port->dev;
1082
1083	if (!host->driver) {
1084		dev_WARN_ONCE(&port->dev, 1, "dport:%s bad devm context\n",
1085			      dev_name(dport_dev));
1086		return ERR_PTR(-ENXIO);
1087	}
1088
1089	if (snprintf(link_name, CXL_TARGET_STRLEN, "dport%d", port_id) >=
1090	    CXL_TARGET_STRLEN)
1091		return ERR_PTR(-EINVAL);
1092
1093	dport = devm_kzalloc(host, sizeof(*dport), GFP_KERNEL);
1094	if (!dport)
1095		return ERR_PTR(-ENOMEM);
1096
1097	dport->dport_dev = dport_dev;
1098	dport->port_id = port_id;
1099	dport->port = port;
1100
1101	if (rcrb == CXL_RESOURCE_NONE) {
1102		rc = cxl_dport_setup_regs(&port->dev, dport,
1103					  component_reg_phys);
1104		if (rc)
1105			return ERR_PTR(rc);
1106	} else {
1107		dport->rcrb.base = rcrb;
1108		component_reg_phys = __rcrb_to_component(dport_dev, &dport->rcrb,
1109							 CXL_RCRB_DOWNSTREAM);
1110		if (component_reg_phys == CXL_RESOURCE_NONE) {
1111			dev_warn(dport_dev, "Invalid Component Registers in RCRB");
1112			return ERR_PTR(-ENXIO);
1113		}
1114
1115		/*
1116		 * RCH @dport is not ready to map until associated with its
1117		 * memdev
1118		 */
1119		rc = cxl_dport_setup_regs(NULL, dport, component_reg_phys);
1120		if (rc)
1121			return ERR_PTR(rc);
1122
1123		dport->rch = true;
1124	}
1125
1126	if (component_reg_phys != CXL_RESOURCE_NONE)
1127		dev_dbg(dport_dev, "Component Registers found for dport: %pa\n",
1128			&component_reg_phys);
1129
1130	cond_cxl_root_lock(port);
1131	rc = add_dport(port, dport);
1132	cond_cxl_root_unlock(port);
1133	if (rc)
1134		return ERR_PTR(rc);
1135
1136	get_device(dport_dev);
1137	rc = devm_add_action_or_reset(host, cxl_dport_remove, dport);
1138	if (rc)
1139		return ERR_PTR(rc);
1140
1141	rc = sysfs_create_link(&port->dev.kobj, &dport_dev->kobj, link_name);
1142	if (rc)
1143		return ERR_PTR(rc);
1144
1145	rc = devm_add_action_or_reset(host, cxl_dport_unlink, dport);
1146	if (rc)
1147		return ERR_PTR(rc);
1148
1149	if (dev_is_pci(dport_dev))
1150		dport->link_latency = cxl_pci_get_latency(to_pci_dev(dport_dev));
1151
1152	return dport;
1153}
1154
1155/**
1156 * devm_cxl_add_dport - append VH downstream port data to a cxl_port
1157 * @port: the cxl_port that references this dport
1158 * @dport_dev: firmware or PCI device representing the dport
1159 * @port_id: identifier for this dport in a decoder's target list
1160 * @component_reg_phys: optional location of CXL component registers
1161 *
1162 * Note that dports are appended to the devm release action's of the
1163 * either the port's host (for root ports), or the port itself (for
1164 * switch ports)
1165 */
1166struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
1167				     struct device *dport_dev, int port_id,
1168				     resource_size_t component_reg_phys)
1169{
1170	struct cxl_dport *dport;
1171
1172	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
1173				     component_reg_phys, CXL_RESOURCE_NONE);
1174	if (IS_ERR(dport)) {
1175		dev_dbg(dport_dev, "failed to add dport to %s: %ld\n",
1176			dev_name(&port->dev), PTR_ERR(dport));
1177	} else {
1178		dev_dbg(dport_dev, "dport added to %s\n",
1179			dev_name(&port->dev));
1180	}
1181
1182	return dport;
1183}
1184EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
1185
1186/**
1187 * devm_cxl_add_rch_dport - append RCH downstream port data to a cxl_port
1188 * @port: the cxl_port that references this dport
1189 * @dport_dev: firmware or PCI device representing the dport
1190 * @port_id: identifier for this dport in a decoder's target list
1191 * @rcrb: mandatory location of a Root Complex Register Block
1192 *
1193 * See CXL 3.0 9.11.8 CXL Devices Attached to an RCH
1194 */
1195struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
1196					 struct device *dport_dev, int port_id,
1197					 resource_size_t rcrb)
1198{
1199	struct cxl_dport *dport;
1200
1201	if (rcrb == CXL_RESOURCE_NONE) {
1202		dev_dbg(&port->dev, "failed to add RCH dport, missing RCRB\n");
1203		return ERR_PTR(-EINVAL);
1204	}
1205
1206	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
1207				     CXL_RESOURCE_NONE, rcrb);
1208	if (IS_ERR(dport)) {
1209		dev_dbg(dport_dev, "failed to add RCH dport to %s: %ld\n",
1210			dev_name(&port->dev), PTR_ERR(dport));
1211	} else {
1212		dev_dbg(dport_dev, "RCH dport added to %s\n",
1213			dev_name(&port->dev));
1214	}
1215
1216	return dport;
1217}
1218EXPORT_SYMBOL_NS_GPL(devm_cxl_add_rch_dport, CXL);
1219
1220static int add_ep(struct cxl_ep *new)
1221{
1222	struct cxl_port *port = new->dport->port;
1223	int rc;
1224
1225	device_lock(&port->dev);
1226	if (port->dead) {
1227		device_unlock(&port->dev);
1228		return -ENXIO;
1229	}
1230	rc = xa_insert(&port->endpoints, (unsigned long)new->ep, new,
1231		       GFP_KERNEL);
1232	device_unlock(&port->dev);
1233
1234	return rc;
1235}
1236
1237/**
1238 * cxl_add_ep - register an endpoint's interest in a port
1239 * @dport: the dport that routes to @ep_dev
1240 * @ep_dev: device representing the endpoint
1241 *
1242 * Intermediate CXL ports are scanned based on the arrival of endpoints.
1243 * When those endpoints depart the port can be destroyed once all
1244 * endpoints that care about that port have been removed.
1245 */
1246static int cxl_add_ep(struct cxl_dport *dport, struct device *ep_dev)
1247{
1248	struct cxl_ep *ep;
1249	int rc;
1250
1251	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
1252	if (!ep)
1253		return -ENOMEM;
1254
1255	ep->ep = get_device(ep_dev);
1256	ep->dport = dport;
1257
1258	rc = add_ep(ep);
1259	if (rc)
1260		cxl_ep_release(ep);
1261	return rc;
1262}
1263
1264struct cxl_find_port_ctx {
1265	const struct device *dport_dev;
1266	const struct cxl_port *parent_port;
1267	struct cxl_dport **dport;
1268};
1269
1270static int match_port_by_dport(struct device *dev, const void *data)
1271{
1272	const struct cxl_find_port_ctx *ctx = data;
1273	struct cxl_dport *dport;
1274	struct cxl_port *port;
1275
1276	if (!is_cxl_port(dev))
1277		return 0;
1278	if (ctx->parent_port && dev->parent != &ctx->parent_port->dev)
1279		return 0;
1280
1281	port = to_cxl_port(dev);
1282	dport = cxl_find_dport_by_dev(port, ctx->dport_dev);
1283	if (ctx->dport)
1284		*ctx->dport = dport;
1285	return dport != NULL;
1286}
1287
1288static struct cxl_port *__find_cxl_port(struct cxl_find_port_ctx *ctx)
1289{
1290	struct device *dev;
1291
1292	if (!ctx->dport_dev)
1293		return NULL;
1294
1295	dev = bus_find_device(&cxl_bus_type, NULL, ctx, match_port_by_dport);
1296	if (dev)
1297		return to_cxl_port(dev);
1298	return NULL;
1299}
1300
1301static struct cxl_port *find_cxl_port(struct device *dport_dev,
1302				      struct cxl_dport **dport)
1303{
1304	struct cxl_find_port_ctx ctx = {
1305		.dport_dev = dport_dev,
1306		.dport = dport,
1307	};
1308	struct cxl_port *port;
1309
1310	port = __find_cxl_port(&ctx);
1311	return port;
1312}
1313
1314static struct cxl_port *find_cxl_port_at(struct cxl_port *parent_port,
1315					 struct device *dport_dev,
1316					 struct cxl_dport **dport)
1317{
1318	struct cxl_find_port_ctx ctx = {
1319		.dport_dev = dport_dev,
1320		.parent_port = parent_port,
1321		.dport = dport,
1322	};
1323	struct cxl_port *port;
1324
1325	port = __find_cxl_port(&ctx);
1326	return port;
1327}
1328
1329/*
1330 * All users of grandparent() are using it to walk PCIe-like switch port
1331 * hierarchy. A PCIe switch is comprised of a bridge device representing the
1332 * upstream switch port and N bridges representing downstream switch ports. When
1333 * bridges stack the grand-parent of a downstream switch port is another
1334 * downstream switch port in the immediate ancestor switch.
1335 */
1336static struct device *grandparent(struct device *dev)
1337{
1338	if (dev && dev->parent)
1339		return dev->parent->parent;
1340	return NULL;
1341}
1342
1343static struct device *endpoint_host(struct cxl_port *endpoint)
1344{
1345	struct cxl_port *port = to_cxl_port(endpoint->dev.parent);
1346
1347	if (is_cxl_root(port))
1348		return port->uport_dev;
1349	return &port->dev;
1350}
1351
1352static void delete_endpoint(void *data)
1353{
1354	struct cxl_memdev *cxlmd = data;
1355	struct cxl_port *endpoint = cxlmd->endpoint;
1356	struct device *host = endpoint_host(endpoint);
1357
1358	device_lock(host);
1359	if (host->driver && !endpoint->dead) {
1360		devm_release_action(host, cxl_unlink_parent_dport, endpoint);
1361		devm_release_action(host, cxl_unlink_uport, endpoint);
1362		devm_release_action(host, unregister_port, endpoint);
1363	}
1364	cxlmd->endpoint = NULL;
1365	device_unlock(host);
1366	put_device(&endpoint->dev);
1367	put_device(host);
1368}
1369
1370int cxl_endpoint_autoremove(struct cxl_memdev *cxlmd, struct cxl_port *endpoint)
1371{
1372	struct device *host = endpoint_host(endpoint);
1373	struct device *dev = &cxlmd->dev;
1374
1375	get_device(host);
1376	get_device(&endpoint->dev);
1377	cxlmd->endpoint = endpoint;
1378	cxlmd->depth = endpoint->depth;
1379	return devm_add_action_or_reset(dev, delete_endpoint, cxlmd);
1380}
1381EXPORT_SYMBOL_NS_GPL(cxl_endpoint_autoremove, CXL);
1382
1383/*
1384 * The natural end of life of a non-root 'cxl_port' is when its parent port goes
1385 * through a ->remove() event ("top-down" unregistration). The unnatural trigger
1386 * for a port to be unregistered is when all memdevs beneath that port have gone
1387 * through ->remove(). This "bottom-up" removal selectively removes individual
1388 * child ports manually. This depends on devm_cxl_add_port() to not change is
1389 * devm action registration order, and for dports to have already been
1390 * destroyed by reap_dports().
1391 */
1392static void delete_switch_port(struct cxl_port *port)
1393{
1394	devm_release_action(port->dev.parent, cxl_unlink_parent_dport, port);
1395	devm_release_action(port->dev.parent, cxl_unlink_uport, port);
1396	devm_release_action(port->dev.parent, unregister_port, port);
1397}
1398
1399static void reap_dports(struct cxl_port *port)
1400{
1401	struct cxl_dport *dport;
1402	unsigned long index;
1403
1404	device_lock_assert(&port->dev);
1405
1406	xa_for_each(&port->dports, index, dport) {
1407		devm_release_action(&port->dev, cxl_dport_unlink, dport);
1408		devm_release_action(&port->dev, cxl_dport_remove, dport);
1409		devm_kfree(&port->dev, dport);
1410	}
1411}
1412
1413struct detach_ctx {
1414	struct cxl_memdev *cxlmd;
1415	int depth;
1416};
1417
1418static int port_has_memdev(struct device *dev, const void *data)
1419{
1420	const struct detach_ctx *ctx = data;
1421	struct cxl_port *port;
1422
1423	if (!is_cxl_port(dev))
1424		return 0;
1425
1426	port = to_cxl_port(dev);
1427	if (port->depth != ctx->depth)
1428		return 0;
1429
1430	return !!cxl_ep_load(port, ctx->cxlmd);
1431}
1432
1433static void cxl_detach_ep(void *data)
1434{
1435	struct cxl_memdev *cxlmd = data;
1436
1437	for (int i = cxlmd->depth - 1; i >= 1; i--) {
1438		struct cxl_port *port, *parent_port;
1439		struct detach_ctx ctx = {
1440			.cxlmd = cxlmd,
1441			.depth = i,
1442		};
1443		struct device *dev;
1444		struct cxl_ep *ep;
1445		bool died = false;
1446
1447		dev = bus_find_device(&cxl_bus_type, NULL, &ctx,
1448				      port_has_memdev);
1449		if (!dev)
1450			continue;
1451		port = to_cxl_port(dev);
1452
1453		parent_port = to_cxl_port(port->dev.parent);
1454		device_lock(&parent_port->dev);
1455		device_lock(&port->dev);
1456		ep = cxl_ep_load(port, cxlmd);
1457		dev_dbg(&cxlmd->dev, "disconnect %s from %s\n",
1458			ep ? dev_name(ep->ep) : "", dev_name(&port->dev));
1459		cxl_ep_remove(port, ep);
1460		if (ep && !port->dead && xa_empty(&port->endpoints) &&
1461		    !is_cxl_root(parent_port) && parent_port->dev.driver) {
1462			/*
1463			 * This was the last ep attached to a dynamically
1464			 * enumerated port. Block new cxl_add_ep() and garbage
1465			 * collect the port.
1466			 */
1467			died = true;
1468			port->dead = true;
1469			reap_dports(port);
1470		}
1471		device_unlock(&port->dev);
1472
1473		if (died) {
1474			dev_dbg(&cxlmd->dev, "delete %s\n",
1475				dev_name(&port->dev));
1476			delete_switch_port(port);
1477		}
1478		put_device(&port->dev);
1479		device_unlock(&parent_port->dev);
1480	}
1481}
1482
1483static resource_size_t find_component_registers(struct device *dev)
1484{
1485	struct cxl_register_map map;
1486	struct pci_dev *pdev;
1487
1488	/*
1489	 * Theoretically, CXL component registers can be hosted on a
1490	 * non-PCI device, in practice, only cxl_test hits this case.
1491	 */
1492	if (!dev_is_pci(dev))
1493		return CXL_RESOURCE_NONE;
1494
1495	pdev = to_pci_dev(dev);
1496
1497	cxl_find_regblock(pdev, CXL_REGLOC_RBI_COMPONENT, &map);
1498	return map.resource;
1499}
1500
1501static int add_port_attach_ep(struct cxl_memdev *cxlmd,
1502			      struct device *uport_dev,
1503			      struct device *dport_dev)
1504{
1505	struct device *dparent = grandparent(dport_dev);
1506	struct cxl_port *port, *parent_port = NULL;
1507	struct cxl_dport *dport, *parent_dport;
1508	resource_size_t component_reg_phys;
1509	int rc;
1510
1511	if (!dparent) {
1512		/*
1513		 * The iteration reached the topology root without finding the
1514		 * CXL-root 'cxl_port' on a previous iteration, fail for now to
1515		 * be re-probed after platform driver attaches.
1516		 */
1517		dev_dbg(&cxlmd->dev, "%s is a root dport\n",
1518			dev_name(dport_dev));
1519		return -ENXIO;
1520	}
1521
1522	parent_port = find_cxl_port(dparent, &parent_dport);
1523	if (!parent_port) {
1524		/* iterate to create this parent_port */
1525		return -EAGAIN;
1526	}
1527
1528	device_lock(&parent_port->dev);
1529	if (!parent_port->dev.driver) {
1530		dev_warn(&cxlmd->dev,
1531			 "port %s:%s disabled, failed to enumerate CXL.mem\n",
1532			 dev_name(&parent_port->dev), dev_name(uport_dev));
1533		port = ERR_PTR(-ENXIO);
1534		goto out;
1535	}
1536
1537	port = find_cxl_port_at(parent_port, dport_dev, &dport);
1538	if (!port) {
1539		component_reg_phys = find_component_registers(uport_dev);
1540		port = devm_cxl_add_port(&parent_port->dev, uport_dev,
1541					 component_reg_phys, parent_dport);
1542		/* retry find to pick up the new dport information */
1543		if (!IS_ERR(port))
1544			port = find_cxl_port_at(parent_port, dport_dev, &dport);
1545	}
1546out:
1547	device_unlock(&parent_port->dev);
1548
1549	if (IS_ERR(port))
1550		rc = PTR_ERR(port);
1551	else {
1552		dev_dbg(&cxlmd->dev, "add to new port %s:%s\n",
1553			dev_name(&port->dev), dev_name(port->uport_dev));
1554		rc = cxl_add_ep(dport, &cxlmd->dev);
1555		if (rc == -EBUSY) {
1556			/*
1557			 * "can't" happen, but this error code means
1558			 * something to the caller, so translate it.
1559			 */
1560			rc = -ENXIO;
1561		}
1562		put_device(&port->dev);
1563	}
1564
1565	put_device(&parent_port->dev);
1566	return rc;
1567}
1568
1569int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
1570{
1571	struct device *dev = &cxlmd->dev;
1572	struct device *iter;
1573	int rc;
1574
1575	/*
1576	 * Skip intermediate port enumeration in the RCH case, there
1577	 * are no ports in between a host bridge and an endpoint.
1578	 */
1579	if (cxlmd->cxlds->rcd)
1580		return 0;
1581
1582	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
1583	if (rc)
1584		return rc;
1585
1586	/*
1587	 * Scan for and add all cxl_ports in this device's ancestry.
1588	 * Repeat until no more ports are added. Abort if a port add
1589	 * attempt fails.
1590	 */
1591retry:
1592	for (iter = dev; iter; iter = grandparent(iter)) {
1593		struct device *dport_dev = grandparent(iter);
1594		struct device *uport_dev;
1595		struct cxl_dport *dport;
1596		struct cxl_port *port;
1597
1598		/*
1599		 * The terminal "grandparent" in PCI is NULL and @platform_bus
1600		 * for platform devices
1601		 */
1602		if (!dport_dev || dport_dev == &platform_bus)
1603			return 0;
1604
1605		uport_dev = dport_dev->parent;
1606		if (!uport_dev) {
1607			dev_warn(dev, "at %s no parent for dport: %s\n",
1608				 dev_name(iter), dev_name(dport_dev));
1609			return -ENXIO;
1610		}
1611
1612		dev_dbg(dev, "scan: iter: %s dport_dev: %s parent: %s\n",
1613			dev_name(iter), dev_name(dport_dev),
1614			dev_name(uport_dev));
1615		port = find_cxl_port(dport_dev, &dport);
1616		if (port) {
1617			dev_dbg(&cxlmd->dev,
1618				"found already registered port %s:%s\n",
1619				dev_name(&port->dev),
1620				dev_name(port->uport_dev));
1621			rc = cxl_add_ep(dport, &cxlmd->dev);
1622
1623			/*
1624			 * If the endpoint already exists in the port's list,
1625			 * that's ok, it was added on a previous pass.
1626			 * Otherwise, retry in add_port_attach_ep() after taking
1627			 * the parent_port lock as the current port may be being
1628			 * reaped.
1629			 */
1630			if (rc && rc != -EBUSY) {
1631				put_device(&port->dev);
1632				return rc;
1633			}
1634
1635			/* Any more ports to add between this one and the root? */
1636			if (!dev_is_cxl_root_child(&port->dev)) {
1637				put_device(&port->dev);
1638				continue;
1639			}
1640
1641			put_device(&port->dev);
1642			return 0;
1643		}
1644
1645		rc = add_port_attach_ep(cxlmd, uport_dev, dport_dev);
1646		/* port missing, try to add parent */
1647		if (rc == -EAGAIN)
1648			continue;
1649		/* failed to add ep or port */
1650		if (rc)
1651			return rc;
1652		/* port added, new descendants possible, start over */
1653		goto retry;
1654	}
1655
1656	return 0;
1657}
1658EXPORT_SYMBOL_NS_GPL(devm_cxl_enumerate_ports, CXL);
1659
1660struct cxl_port *cxl_pci_find_port(struct pci_dev *pdev,
1661				   struct cxl_dport **dport)
1662{
1663	return find_cxl_port(pdev->dev.parent, dport);
1664}
1665EXPORT_SYMBOL_NS_GPL(cxl_pci_find_port, CXL);
1666
1667struct cxl_port *cxl_mem_find_port(struct cxl_memdev *cxlmd,
1668				   struct cxl_dport **dport)
1669{
1670	return find_cxl_port(grandparent(&cxlmd->dev), dport);
1671}
1672EXPORT_SYMBOL_NS_GPL(cxl_mem_find_port, CXL);
1673
1674static int decoder_populate_targets(struct cxl_switch_decoder *cxlsd,
1675				    struct cxl_port *port, int *target_map)
1676{
1677	int i;
1678
1679	if (!target_map)
1680		return 0;
1681
1682	device_lock_assert(&port->dev);
1683
1684	if (xa_empty(&port->dports))
1685		return -EINVAL;
1686
1687	guard(rwsem_write)(&cxl_region_rwsem);
1688	for (i = 0; i < cxlsd->cxld.interleave_ways; i++) {
1689		struct cxl_dport *dport = find_dport(port, target_map[i]);
1690
1691		if (!dport)
1692			return -ENXIO;
1693		cxlsd->target[i] = dport;
1694	}
1695
1696	return 0;
1697}
1698
1699struct cxl_dport *cxl_hb_modulo(struct cxl_root_decoder *cxlrd, int pos)
1700{
1701	struct cxl_switch_decoder *cxlsd = &cxlrd->cxlsd;
1702	struct cxl_decoder *cxld = &cxlsd->cxld;
1703	int iw;
1704
1705	iw = cxld->interleave_ways;
1706	if (dev_WARN_ONCE(&cxld->dev, iw != cxlsd->nr_targets,
1707			  "misconfigured root decoder\n"))
1708		return NULL;
1709
1710	return cxlrd->cxlsd.target[pos % iw];
1711}
1712EXPORT_SYMBOL_NS_GPL(cxl_hb_modulo, CXL);
1713
1714static struct lock_class_key cxl_decoder_key;
1715
1716/**
1717 * cxl_decoder_init - Common decoder setup / initialization
1718 * @port: owning port of this decoder
1719 * @cxld: common decoder properties to initialize
1720 *
1721 * A port may contain one or more decoders. Each of those decoders
1722 * enable some address space for CXL.mem utilization. A decoder is
1723 * expected to be configured by the caller before registering via
1724 * cxl_decoder_add()
1725 */
1726static int cxl_decoder_init(struct cxl_port *port, struct cxl_decoder *cxld)
1727{
1728	struct device *dev;
1729	int rc;
1730
1731	rc = ida_alloc(&port->decoder_ida, GFP_KERNEL);
1732	if (rc < 0)
1733		return rc;
1734
1735	/* need parent to stick around to release the id */
1736	get_device(&port->dev);
1737	cxld->id = rc;
1738
1739	dev = &cxld->dev;
1740	device_initialize(dev);
1741	lockdep_set_class(&dev->mutex, &cxl_decoder_key);
1742	device_set_pm_not_required(dev);
1743	dev->parent = &port->dev;
1744	dev->bus = &cxl_bus_type;
1745
1746	/* Pre initialize an "empty" decoder */
1747	cxld->interleave_ways = 1;
1748	cxld->interleave_granularity = PAGE_SIZE;
1749	cxld->target_type = CXL_DECODER_HOSTONLYMEM;
1750	cxld->hpa_range = (struct range) {
1751		.start = 0,
1752		.end = -1,
1753	};
1754
1755	return 0;
1756}
1757
1758static int cxl_switch_decoder_init(struct cxl_port *port,
1759				   struct cxl_switch_decoder *cxlsd,
1760				   int nr_targets)
1761{
1762	if (nr_targets > CXL_DECODER_MAX_INTERLEAVE)
1763		return -EINVAL;
1764
1765	cxlsd->nr_targets = nr_targets;
1766	return cxl_decoder_init(port, &cxlsd->cxld);
1767}
1768
1769/**
1770 * cxl_root_decoder_alloc - Allocate a root level decoder
1771 * @port: owning CXL root of this decoder
1772 * @nr_targets: static number of downstream targets
1773 * @calc_hb: which host bridge covers the n'th position by granularity
1774 *
1775 * Return: A new cxl decoder to be registered by cxl_decoder_add(). A
1776 * 'CXL root' decoder is one that decodes from a top-level / static platform
1777 * firmware description of CXL resources into a CXL standard decode
1778 * topology.
1779 */
1780struct cxl_root_decoder *cxl_root_decoder_alloc(struct cxl_port *port,
1781						unsigned int nr_targets,
1782						cxl_calc_hb_fn calc_hb)
1783{
1784	struct cxl_root_decoder *cxlrd;
1785	struct cxl_switch_decoder *cxlsd;
1786	struct cxl_decoder *cxld;
1787	int rc;
1788
1789	if (!is_cxl_root(port))
1790		return ERR_PTR(-EINVAL);
1791
1792	cxlrd = kzalloc(struct_size(cxlrd, cxlsd.target, nr_targets),
1793			GFP_KERNEL);
1794	if (!cxlrd)
1795		return ERR_PTR(-ENOMEM);
1796
1797	cxlsd = &cxlrd->cxlsd;
1798	rc = cxl_switch_decoder_init(port, cxlsd, nr_targets);
1799	if (rc) {
1800		kfree(cxlrd);
1801		return ERR_PTR(rc);
1802	}
1803
1804	cxlrd->calc_hb = calc_hb;
1805	mutex_init(&cxlrd->range_lock);
1806
1807	cxld = &cxlsd->cxld;
1808	cxld->dev.type = &cxl_decoder_root_type;
1809	/*
1810	 * cxl_root_decoder_release() special cases negative ids to
1811	 * detect memregion_alloc() failures.
1812	 */
1813	atomic_set(&cxlrd->region_id, -1);
1814	rc = memregion_alloc(GFP_KERNEL);
1815	if (rc < 0) {
1816		put_device(&cxld->dev);
1817		return ERR_PTR(rc);
1818	}
1819
1820	atomic_set(&cxlrd->region_id, rc);
1821	cxlrd->qos_class = CXL_QOS_CLASS_INVALID;
1822	return cxlrd;
1823}
1824EXPORT_SYMBOL_NS_GPL(cxl_root_decoder_alloc, CXL);
1825
1826/**
1827 * cxl_switch_decoder_alloc - Allocate a switch level decoder
1828 * @port: owning CXL switch port of this decoder
1829 * @nr_targets: max number of dynamically addressable downstream targets
1830 *
1831 * Return: A new cxl decoder to be registered by cxl_decoder_add(). A
1832 * 'switch' decoder is any decoder that can be enumerated by PCIe
1833 * topology and the HDM Decoder Capability. This includes the decoders
1834 * that sit between Switch Upstream Ports / Switch Downstream Ports and
1835 * Host Bridges / Root Ports.
1836 */
1837struct cxl_switch_decoder *cxl_switch_decoder_alloc(struct cxl_port *port,
1838						    unsigned int nr_targets)
1839{
1840	struct cxl_switch_decoder *cxlsd;
1841	struct cxl_decoder *cxld;
1842	int rc;
1843
1844	if (is_cxl_root(port) || is_cxl_endpoint(port))
1845		return ERR_PTR(-EINVAL);
1846
1847	cxlsd = kzalloc(struct_size(cxlsd, target, nr_targets), GFP_KERNEL);
1848	if (!cxlsd)
1849		return ERR_PTR(-ENOMEM);
1850
1851	rc = cxl_switch_decoder_init(port, cxlsd, nr_targets);
1852	if (rc) {
1853		kfree(cxlsd);
1854		return ERR_PTR(rc);
1855	}
1856
1857	cxld = &cxlsd->cxld;
1858	cxld->dev.type = &cxl_decoder_switch_type;
1859	return cxlsd;
1860}
1861EXPORT_SYMBOL_NS_GPL(cxl_switch_decoder_alloc, CXL);
1862
1863/**
1864 * cxl_endpoint_decoder_alloc - Allocate an endpoint decoder
1865 * @port: owning port of this decoder
1866 *
1867 * Return: A new cxl decoder to be registered by cxl_decoder_add()
1868 */
1869struct cxl_endpoint_decoder *cxl_endpoint_decoder_alloc(struct cxl_port *port)
1870{
1871	struct cxl_endpoint_decoder *cxled;
1872	struct cxl_decoder *cxld;
1873	int rc;
1874
1875	if (!is_cxl_endpoint(port))
1876		return ERR_PTR(-EINVAL);
1877
1878	cxled = kzalloc(sizeof(*cxled), GFP_KERNEL);
1879	if (!cxled)
1880		return ERR_PTR(-ENOMEM);
1881
1882	cxled->pos = -1;
1883	cxld = &cxled->cxld;
1884	rc = cxl_decoder_init(port, cxld);
1885	if (rc)	 {
1886		kfree(cxled);
1887		return ERR_PTR(rc);
1888	}
1889
1890	cxld->dev.type = &cxl_decoder_endpoint_type;
1891	return cxled;
1892}
1893EXPORT_SYMBOL_NS_GPL(cxl_endpoint_decoder_alloc, CXL);
1894
1895/**
1896 * cxl_decoder_add_locked - Add a decoder with targets
1897 * @cxld: The cxl decoder allocated by cxl_<type>_decoder_alloc()
1898 * @target_map: A list of downstream ports that this decoder can direct memory
1899 *              traffic to. These numbers should correspond with the port number
1900 *              in the PCIe Link Capabilities structure.
1901 *
1902 * Certain types of decoders may not have any targets. The main example of this
1903 * is an endpoint device. A more awkward example is a hostbridge whose root
1904 * ports get hot added (technically possible, though unlikely).
1905 *
1906 * This is the locked variant of cxl_decoder_add().
1907 *
1908 * Context: Process context. Expects the device lock of the port that owns the
1909 *	    @cxld to be held.
1910 *
1911 * Return: Negative error code if the decoder wasn't properly configured; else
1912 *	   returns 0.
1913 */
1914int cxl_decoder_add_locked(struct cxl_decoder *cxld, int *target_map)
1915{
1916	struct cxl_port *port;
1917	struct device *dev;
1918	int rc;
1919
1920	if (WARN_ON_ONCE(!cxld))
1921		return -EINVAL;
1922
1923	if (WARN_ON_ONCE(IS_ERR(cxld)))
1924		return PTR_ERR(cxld);
1925
1926	if (cxld->interleave_ways < 1)
1927		return -EINVAL;
1928
1929	dev = &cxld->dev;
1930
1931	port = to_cxl_port(cxld->dev.parent);
1932	if (!is_endpoint_decoder(dev)) {
1933		struct cxl_switch_decoder *cxlsd = to_cxl_switch_decoder(dev);
1934
1935		rc = decoder_populate_targets(cxlsd, port, target_map);
1936		if (rc && (cxld->flags & CXL_DECODER_F_ENABLE)) {
1937			dev_err(&port->dev,
1938				"Failed to populate active decoder targets\n");
1939			return rc;
1940		}
1941	}
1942
1943	rc = dev_set_name(dev, "decoder%d.%d", port->id, cxld->id);
1944	if (rc)
1945		return rc;
1946
1947	return device_add(dev);
1948}
1949EXPORT_SYMBOL_NS_GPL(cxl_decoder_add_locked, CXL);
1950
1951/**
1952 * cxl_decoder_add - Add a decoder with targets
1953 * @cxld: The cxl decoder allocated by cxl_<type>_decoder_alloc()
1954 * @target_map: A list of downstream ports that this decoder can direct memory
1955 *              traffic to. These numbers should correspond with the port number
1956 *              in the PCIe Link Capabilities structure.
1957 *
1958 * This is the unlocked variant of cxl_decoder_add_locked().
1959 * See cxl_decoder_add_locked().
1960 *
1961 * Context: Process context. Takes and releases the device lock of the port that
1962 *	    owns the @cxld.
1963 */
1964int cxl_decoder_add(struct cxl_decoder *cxld, int *target_map)
1965{
1966	struct cxl_port *port;
1967	int rc;
1968
1969	if (WARN_ON_ONCE(!cxld))
1970		return -EINVAL;
1971
1972	if (WARN_ON_ONCE(IS_ERR(cxld)))
1973		return PTR_ERR(cxld);
1974
1975	port = to_cxl_port(cxld->dev.parent);
1976
1977	device_lock(&port->dev);
1978	rc = cxl_decoder_add_locked(cxld, target_map);
1979	device_unlock(&port->dev);
1980
1981	return rc;
1982}
1983EXPORT_SYMBOL_NS_GPL(cxl_decoder_add, CXL);
1984
1985static void cxld_unregister(void *dev)
1986{
1987	struct cxl_endpoint_decoder *cxled;
1988
1989	if (is_endpoint_decoder(dev)) {
1990		cxled = to_cxl_endpoint_decoder(dev);
1991		cxl_decoder_kill_region(cxled);
1992	}
1993
1994	device_unregister(dev);
1995}
1996
1997int cxl_decoder_autoremove(struct device *host, struct cxl_decoder *cxld)
1998{
1999	return devm_add_action_or_reset(host, cxld_unregister, &cxld->dev);
2000}
2001EXPORT_SYMBOL_NS_GPL(cxl_decoder_autoremove, CXL);
2002
2003/**
2004 * __cxl_driver_register - register a driver for the cxl bus
2005 * @cxl_drv: cxl driver structure to attach
2006 * @owner: owning module/driver
2007 * @modname: KBUILD_MODNAME for parent driver
2008 */
2009int __cxl_driver_register(struct cxl_driver *cxl_drv, struct module *owner,
2010			  const char *modname)
2011{
2012	if (!cxl_drv->probe) {
2013		pr_debug("%s ->probe() must be specified\n", modname);
2014		return -EINVAL;
2015	}
2016
2017	if (!cxl_drv->name) {
2018		pr_debug("%s ->name must be specified\n", modname);
2019		return -EINVAL;
2020	}
2021
2022	if (!cxl_drv->id) {
2023		pr_debug("%s ->id must be specified\n", modname);
2024		return -EINVAL;
2025	}
2026
2027	cxl_drv->drv.bus = &cxl_bus_type;
2028	cxl_drv->drv.owner = owner;
2029	cxl_drv->drv.mod_name = modname;
2030	cxl_drv->drv.name = cxl_drv->name;
2031
2032	return driver_register(&cxl_drv->drv);
2033}
2034EXPORT_SYMBOL_NS_GPL(__cxl_driver_register, CXL);
2035
2036void cxl_driver_unregister(struct cxl_driver *cxl_drv)
2037{
2038	driver_unregister(&cxl_drv->drv);
2039}
2040EXPORT_SYMBOL_NS_GPL(cxl_driver_unregister, CXL);
2041
2042static int cxl_bus_uevent(const struct device *dev, struct kobj_uevent_env *env)
2043{
2044	return add_uevent_var(env, "MODALIAS=" CXL_MODALIAS_FMT,
2045			      cxl_device_id(dev));
2046}
2047
2048static int cxl_bus_match(struct device *dev, struct device_driver *drv)
2049{
2050	return cxl_device_id(dev) == to_cxl_drv(drv)->id;
2051}
2052
2053static int cxl_bus_probe(struct device *dev)
2054{
2055	int rc;
2056
2057	rc = to_cxl_drv(dev->driver)->probe(dev);
2058	dev_dbg(dev, "probe: %d\n", rc);
2059	return rc;
2060}
2061
2062static void cxl_bus_remove(struct device *dev)
2063{
2064	struct cxl_driver *cxl_drv = to_cxl_drv(dev->driver);
2065
2066	if (cxl_drv->remove)
2067		cxl_drv->remove(dev);
2068}
2069
2070static struct workqueue_struct *cxl_bus_wq;
2071
2072static void cxl_bus_rescan_queue(struct work_struct *w)
2073{
2074	int rc = bus_rescan_devices(&cxl_bus_type);
2075
2076	pr_debug("CXL bus rescan result: %d\n", rc);
2077}
2078
2079void cxl_bus_rescan(void)
2080{
2081	static DECLARE_WORK(rescan_work, cxl_bus_rescan_queue);
2082
2083	queue_work(cxl_bus_wq, &rescan_work);
2084}
2085EXPORT_SYMBOL_NS_GPL(cxl_bus_rescan, CXL);
2086
2087void cxl_bus_drain(void)
2088{
2089	drain_workqueue(cxl_bus_wq);
2090}
2091EXPORT_SYMBOL_NS_GPL(cxl_bus_drain, CXL);
2092
2093bool schedule_cxl_memdev_detach(struct cxl_memdev *cxlmd)
2094{
2095	return queue_work(cxl_bus_wq, &cxlmd->detach_work);
2096}
2097EXPORT_SYMBOL_NS_GPL(schedule_cxl_memdev_detach, CXL);
2098
2099static void combine_coordinates(struct access_coordinate *c1,
2100				struct access_coordinate *c2)
2101{
2102		if (c2->write_bandwidth)
2103			c1->write_bandwidth = min(c1->write_bandwidth,
2104						  c2->write_bandwidth);
2105		c1->write_latency += c2->write_latency;
2106
2107		if (c2->read_bandwidth)
2108			c1->read_bandwidth = min(c1->read_bandwidth,
2109						 c2->read_bandwidth);
2110		c1->read_latency += c2->read_latency;
2111}
2112
2113/**
2114 * cxl_endpoint_get_perf_coordinates - Retrieve performance numbers stored in dports
2115 *				   of CXL path
2116 * @port: endpoint cxl_port
2117 * @coord: output performance data
2118 *
2119 * Return: errno on failure, 0 on success.
2120 */
2121int cxl_endpoint_get_perf_coordinates(struct cxl_port *port,
2122				      struct access_coordinate *coord)
2123{
2124	struct access_coordinate c = {
2125		.read_bandwidth = UINT_MAX,
2126		.write_bandwidth = UINT_MAX,
2127	};
2128	struct cxl_port *iter = port;
2129	struct cxl_dport *dport;
2130	struct pci_dev *pdev;
2131	unsigned int bw;
2132
2133	if (!is_cxl_endpoint(port))
2134		return -EINVAL;
2135
2136	dport = iter->parent_dport;
2137
2138	/*
2139	 * Exit the loop when the parent port of the current port is cxl root.
2140	 * The iterative loop starts at the endpoint and gathers the
2141	 * latency of the CXL link from the current iter to the next downstream
2142	 * port each iteration. If the parent is cxl root then there is
2143	 * nothing to gather.
2144	 */
2145	while (iter && !is_cxl_root(to_cxl_port(iter->dev.parent))) {
2146		combine_coordinates(&c, &dport->sw_coord);
2147		c.write_latency += dport->link_latency;
2148		c.read_latency += dport->link_latency;
2149
2150		iter = to_cxl_port(iter->dev.parent);
2151		dport = iter->parent_dport;
2152	}
2153
2154	/* Augment with the generic port (host bridge) perf data */
2155	combine_coordinates(&c, &dport->hb_coord);
2156
2157	/* Get the calculated PCI paths bandwidth */
2158	pdev = to_pci_dev(port->uport_dev->parent);
2159	bw = pcie_bandwidth_available(pdev, NULL, NULL, NULL);
2160	if (bw == 0)
2161		return -ENXIO;
2162	bw /= BITS_PER_BYTE;
2163
2164	c.write_bandwidth = min(c.write_bandwidth, bw);
2165	c.read_bandwidth = min(c.read_bandwidth, bw);
2166
2167	*coord = c;
2168
2169	return 0;
2170}
2171EXPORT_SYMBOL_NS_GPL(cxl_endpoint_get_perf_coordinates, CXL);
2172
2173/* for user tooling to ensure port disable work has completed */
2174static ssize_t flush_store(const struct bus_type *bus, const char *buf, size_t count)
2175{
2176	if (sysfs_streq(buf, "1")) {
2177		flush_workqueue(cxl_bus_wq);
2178		return count;
2179	}
2180
2181	return -EINVAL;
2182}
2183
2184static BUS_ATTR_WO(flush);
2185
2186static struct attribute *cxl_bus_attributes[] = {
2187	&bus_attr_flush.attr,
2188	NULL,
2189};
2190
2191static struct attribute_group cxl_bus_attribute_group = {
2192	.attrs = cxl_bus_attributes,
2193};
2194
2195static const struct attribute_group *cxl_bus_attribute_groups[] = {
2196	&cxl_bus_attribute_group,
2197	NULL,
2198};
2199
2200struct bus_type cxl_bus_type = {
2201	.name = "cxl",
2202	.uevent = cxl_bus_uevent,
2203	.match = cxl_bus_match,
2204	.probe = cxl_bus_probe,
2205	.remove = cxl_bus_remove,
2206	.bus_groups = cxl_bus_attribute_groups,
2207};
2208EXPORT_SYMBOL_NS_GPL(cxl_bus_type, CXL);
2209
2210static struct dentry *cxl_debugfs;
2211
2212struct dentry *cxl_debugfs_create_dir(const char *dir)
2213{
2214	return debugfs_create_dir(dir, cxl_debugfs);
2215}
2216EXPORT_SYMBOL_NS_GPL(cxl_debugfs_create_dir, CXL);
2217
2218static __init int cxl_core_init(void)
2219{
2220	int rc;
2221
2222	cxl_debugfs = debugfs_create_dir("cxl", NULL);
2223
2224	cxl_mbox_init();
2225
2226	rc = cxl_memdev_init();
2227	if (rc)
2228		return rc;
2229
2230	cxl_bus_wq = alloc_ordered_workqueue("cxl_port", 0);
2231	if (!cxl_bus_wq) {
2232		rc = -ENOMEM;
2233		goto err_wq;
2234	}
2235
2236	rc = bus_register(&cxl_bus_type);
2237	if (rc)
2238		goto err_bus;
2239
2240	rc = cxl_region_init();
2241	if (rc)
2242		goto err_region;
2243
2244	return 0;
2245
2246err_region:
2247	bus_unregister(&cxl_bus_type);
2248err_bus:
2249	destroy_workqueue(cxl_bus_wq);
2250err_wq:
2251	cxl_memdev_exit();
2252	return rc;
2253}
2254
2255static void cxl_core_exit(void)
2256{
2257	cxl_region_exit();
2258	bus_unregister(&cxl_bus_type);
2259	destroy_workqueue(cxl_bus_wq);
2260	cxl_memdev_exit();
2261	debugfs_remove_recursive(cxl_debugfs);
2262}
2263
2264subsys_initcall(cxl_core_init);
2265module_exit(cxl_core_exit);
2266MODULE_LICENSE("GPL v2");
2267MODULE_IMPORT_NS(CXL);