1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Intel(R) Trace Hub Memory Storage Unit
   4 *
   5 * Copyright (C) 2014-2015 Intel Corporation.
   6 */
   7
   8#define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
   9
  10#include <linux/types.h>
  11#include <linux/module.h>
  12#include <linux/device.h>
  13#include <linux/uaccess.h>
  14#include <linux/sizes.h>
  15#include <linux/printk.h>
  16#include <linux/slab.h>
  17#include <linux/mm.h>
  18#include <linux/fs.h>
  19#include <linux/io.h>
  20#include <linux/workqueue.h>
  21#include <linux/dma-mapping.h>
  22
  23#ifdef CONFIG_X86
  24#include <asm/set_memory.h>
  25#endif
  26
  27#include <linux/intel_th.h>
  28#include "intel_th.h"
  29#include "msu.h"
  30
  31#define msc_dev(x) (&(x)->thdev->dev)
  32
  33/*
  34 * Lockout state transitions:
  35 *   READY -> INUSE -+-> LOCKED -+-> READY -> etc.
  36 *                   \-----------/
  37 * WIN_READY:	window can be used by HW
  38 * WIN_INUSE:	window is in use
  39 * WIN_LOCKED:	window is filled up and is being processed by the buffer
  40 * handling code
  41 *
  42 * All state transitions happen automatically, except for the LOCKED->READY,
  43 * which needs to be signalled by the buffer code by calling
  44 * intel_th_msc_window_unlock().
  45 *
  46 * When the interrupt handler has to switch to the next window, it checks
  47 * whether it's READY, and if it is, it performs the switch and tracing
  48 * continues. If it's LOCKED, it stops the trace.
  49 */
  50enum lockout_state {
  51	WIN_READY = 0,
  52	WIN_INUSE,
  53	WIN_LOCKED
  54};
  55
  56/**
  57 * struct msc_window - multiblock mode window descriptor
  58 * @entry:	window list linkage (msc::win_list)
  59 * @pgoff:	page offset into the buffer that this window starts at
  60 * @lockout:	lockout state, see comment below
  61 * @lo_lock:	lockout state serialization
  62 * @nr_blocks:	number of blocks (pages) in this window
  63 * @nr_segs:	number of segments in this window (<= @nr_blocks)
  64 * @_sgt:	array of block descriptors
  65 * @sgt:	array of block descriptors
  66 */
  67struct msc_window {
  68	struct list_head	entry;
  69	unsigned long		pgoff;
  70	enum lockout_state	lockout;
  71	spinlock_t		lo_lock;
  72	unsigned int		nr_blocks;
  73	unsigned int		nr_segs;
  74	struct msc		*msc;
  75	struct sg_table		_sgt;
  76	struct sg_table		*sgt;
  77};
  78
  79/**
  80 * struct msc_iter - iterator for msc buffer
  81 * @entry:		msc::iter_list linkage
  82 * @msc:		pointer to the MSC device
  83 * @start_win:		oldest window
  84 * @win:		current window
  85 * @offset:		current logical offset into the buffer
  86 * @start_block:	oldest block in the window
  87 * @block:		block number in the window
  88 * @block_off:		offset into current block
  89 * @wrap_count:		block wrapping handling
  90 * @eof:		end of buffer reached
  91 */
  92struct msc_iter {
  93	struct list_head	entry;
  94	struct msc		*msc;
  95	struct msc_window	*start_win;
  96	struct msc_window	*win;
  97	unsigned long		offset;
  98	struct scatterlist	*start_block;
  99	struct scatterlist	*block;
 100	unsigned int		block_off;
 101	unsigned int		wrap_count;
 102	unsigned int		eof;
 103};
 104
 105/**
 106 * struct msc - MSC device representation
 107 * @reg_base:		register window base address
 108 * @thdev:		intel_th_device pointer
 109 * @mbuf:		MSU buffer, if assigned
 110 * @mbuf_priv		MSU buffer's private data, if @mbuf
 111 * @win_list:		list of windows in multiblock mode
 112 * @single_sgt:		single mode buffer
 113 * @cur_win:		current window
 114 * @nr_pages:		total number of pages allocated for this buffer
 115 * @single_sz:		amount of data in single mode
 116 * @single_wrap:	single mode wrap occurred
 117 * @base:		buffer's base pointer
 118 * @base_addr:		buffer's base address
 119 * @user_count:		number of users of the buffer
 120 * @mmap_count:		number of mappings
 121 * @buf_mutex:		mutex to serialize access to buffer-related bits
 122
 123 * @enabled:		MSC is enabled
 124 * @wrap:		wrapping is enabled
 125 * @mode:		MSC operating mode
 126 * @burst_len:		write burst length
 127 * @index:		number of this MSC in the MSU
 128 */
 129struct msc {
 130	void __iomem		*reg_base;
 131	void __iomem		*msu_base;
 132	struct intel_th_device	*thdev;
 133
 134	const struct msu_buffer	*mbuf;
 135	void			*mbuf_priv;
 136
 137	struct work_struct	work;
 138	struct list_head	win_list;
 139	struct sg_table		single_sgt;
 140	struct msc_window	*cur_win;
 141	struct msc_window	*switch_on_unlock;
 142	unsigned long		nr_pages;
 143	unsigned long		single_sz;
 144	unsigned int		single_wrap : 1;
 145	void			*base;
 146	dma_addr_t		base_addr;
 147	u32			orig_addr;
 148	u32			orig_sz;
 149
 150	/* <0: no buffer, 0: no users, >0: active users */
 151	atomic_t		user_count;
 152
 153	atomic_t		mmap_count;
 154	struct mutex		buf_mutex;
 155
 156	struct list_head	iter_list;
 157
 158	bool			stop_on_full;
 159
 160	/* config */
 161	unsigned int		enabled : 1,
 162				wrap	: 1,
 163				do_irq	: 1,
 164				multi_is_broken : 1;
 165	unsigned int		mode;
 166	unsigned int		burst_len;
 167	unsigned int		index;
 168};
 169
 170static LIST_HEAD(msu_buffer_list);
 171static DEFINE_MUTEX(msu_buffer_mutex);
 172
 173/**
 174 * struct msu_buffer_entry - internal MSU buffer bookkeeping
 175 * @entry:	link to msu_buffer_list
 176 * @mbuf:	MSU buffer object
 177 * @owner:	module that provides this MSU buffer
 178 */
 179struct msu_buffer_entry {
 180	struct list_head	entry;
 181	const struct msu_buffer	*mbuf;
 182	struct module		*owner;
 183};
 184
 185static struct msu_buffer_entry *__msu_buffer_entry_find(const char *name)
 186{
 187	struct msu_buffer_entry *mbe;
 188
 189	lockdep_assert_held(&msu_buffer_mutex);
 190
 191	list_for_each_entry(mbe, &msu_buffer_list, entry) {
 192		if (!strcmp(mbe->mbuf->name, name))
 193			return mbe;
 194	}
 195
 196	return NULL;
 197}
 198
 199static const struct msu_buffer *
 200msu_buffer_get(const char *name)
 201{
 202	struct msu_buffer_entry *mbe;
 203
 204	mutex_lock(&msu_buffer_mutex);
 205	mbe = __msu_buffer_entry_find(name);
 206	if (mbe && !try_module_get(mbe->owner))
 207		mbe = NULL;
 208	mutex_unlock(&msu_buffer_mutex);
 209
 210	return mbe ? mbe->mbuf : NULL;
 211}
 212
 213static void msu_buffer_put(const struct msu_buffer *mbuf)
 214{
 215	struct msu_buffer_entry *mbe;
 216
 217	mutex_lock(&msu_buffer_mutex);
 218	mbe = __msu_buffer_entry_find(mbuf->name);
 219	if (mbe)
 220		module_put(mbe->owner);
 221	mutex_unlock(&msu_buffer_mutex);
 222}
 223
 224int intel_th_msu_buffer_register(const struct msu_buffer *mbuf,
 225				 struct module *owner)
 226{
 227	struct msu_buffer_entry *mbe;
 228	int ret = 0;
 229
 230	mbe = kzalloc(sizeof(*mbe), GFP_KERNEL);
 231	if (!mbe)
 232		return -ENOMEM;
 233
 234	mutex_lock(&msu_buffer_mutex);
 235	if (__msu_buffer_entry_find(mbuf->name)) {
 236		ret = -EEXIST;
 237		kfree(mbe);
 238		goto unlock;
 239	}
 240
 241	mbe->mbuf = mbuf;
 242	mbe->owner = owner;
 243	list_add_tail(&mbe->entry, &msu_buffer_list);
 244unlock:
 245	mutex_unlock(&msu_buffer_mutex);
 246
 247	return ret;
 248}
 249EXPORT_SYMBOL_GPL(intel_th_msu_buffer_register);
 250
 251void intel_th_msu_buffer_unregister(const struct msu_buffer *mbuf)
 252{
 253	struct msu_buffer_entry *mbe;
 254
 255	mutex_lock(&msu_buffer_mutex);
 256	mbe = __msu_buffer_entry_find(mbuf->name);
 257	if (mbe) {
 258		list_del(&mbe->entry);
 259		kfree(mbe);
 260	}
 261	mutex_unlock(&msu_buffer_mutex);
 262}
 263EXPORT_SYMBOL_GPL(intel_th_msu_buffer_unregister);
 264
 265static inline bool msc_block_is_empty(struct msc_block_desc *bdesc)
 266{
 267	/* header hasn't been written */
 268	if (!bdesc->valid_dw)
 269		return true;
 270
 271	/* valid_dw includes the header */
 272	if (!msc_data_sz(bdesc))
 273		return true;
 274
 275	return false;
 276}
 277
 278static inline struct scatterlist *msc_win_base_sg(struct msc_window *win)
 279{
 280	return win->sgt->sgl;
 281}
 282
 283static inline struct msc_block_desc *msc_win_base(struct msc_window *win)
 284{
 285	return sg_virt(msc_win_base_sg(win));
 286}
 287
 288static inline dma_addr_t msc_win_base_dma(struct msc_window *win)
 289{
 290	return sg_dma_address(msc_win_base_sg(win));
 291}
 292
 293static inline unsigned long
 294msc_win_base_pfn(struct msc_window *win)
 295{
 296	return PFN_DOWN(msc_win_base_dma(win));
 297}
 298
 299/**
 300 * msc_is_last_win() - check if a window is the last one for a given MSC
 301 * @win:	window
 302 * Return:	true if @win is the last window in MSC's multiblock buffer
 303 */
 304static inline bool msc_is_last_win(struct msc_window *win)
 305{
 306	return win->entry.next == &win->msc->win_list;
 307}
 308
 309/**
 310 * msc_next_window() - return next window in the multiblock buffer
 311 * @win:	current window
 312 *
 313 * Return:	window following the current one
 314 */
 315static struct msc_window *msc_next_window(struct msc_window *win)
 316{
 317	if (msc_is_last_win(win))
 318		return list_first_entry(&win->msc->win_list, struct msc_window,
 319					entry);
 320
 321	return list_next_entry(win, entry);
 322}
 323
 324static size_t msc_win_total_sz(struct msc_window *win)
 325{
 326	struct scatterlist *sg;
 327	unsigned int blk;
 328	size_t size = 0;
 329
 330	for_each_sg(win->sgt->sgl, sg, win->nr_segs, blk) {
 331		struct msc_block_desc *bdesc = sg_virt(sg);
 332
 333		if (msc_block_wrapped(bdesc))
 334			return (size_t)win->nr_blocks << PAGE_SHIFT;
 335
 336		size += msc_total_sz(bdesc);
 337		if (msc_block_last_written(bdesc))
 338			break;
 339	}
 340
 341	return size;
 342}
 343
 344/**
 345 * msc_find_window() - find a window matching a given sg_table
 346 * @msc:	MSC device
 347 * @sgt:	SG table of the window
 348 * @nonempty:	skip over empty windows
 349 *
 350 * Return:	MSC window structure pointer or NULL if the window
 351 *		could not be found.
 352 */
 353static struct msc_window *
 354msc_find_window(struct msc *msc, struct sg_table *sgt, bool nonempty)
 355{
 356	struct msc_window *win;
 357	unsigned int found = 0;
 358
 359	if (list_empty(&msc->win_list))
 360		return NULL;
 361
 362	/*
 363	 * we might need a radix tree for this, depending on how
 364	 * many windows a typical user would allocate; ideally it's
 365	 * something like 2, in which case we're good
 366	 */
 367	list_for_each_entry(win, &msc->win_list, entry) {
 368		if (win->sgt == sgt)
 369			found++;
 370
 371		/* skip the empty ones */
 372		if (nonempty && msc_block_is_empty(msc_win_base(win)))
 373			continue;
 374
 375		if (found)
 376			return win;
 377	}
 378
 379	return NULL;
 380}
 381
 382/**
 383 * msc_oldest_window() - locate the window with oldest data
 384 * @msc:	MSC device
 385 *
 386 * This should only be used in multiblock mode. Caller should hold the
 387 * msc::user_count reference.
 388 *
 389 * Return:	the oldest window with valid data
 390 */
 391static struct msc_window *msc_oldest_window(struct msc *msc)
 392{
 393	struct msc_window *win;
 394
 395	if (list_empty(&msc->win_list))
 396		return NULL;
 397
 398	win = msc_find_window(msc, msc_next_window(msc->cur_win)->sgt, true);
 399	if (win)
 400		return win;
 401
 402	return list_first_entry(&msc->win_list, struct msc_window, entry);
 403}
 404
 405/**
 406 * msc_win_oldest_sg() - locate the oldest block in a given window
 407 * @win:	window to look at
 408 *
 409 * Return:	index of the block with the oldest data
 410 */
 411static struct scatterlist *msc_win_oldest_sg(struct msc_window *win)
 412{
 413	unsigned int blk;
 414	struct scatterlist *sg;
 415	struct msc_block_desc *bdesc = msc_win_base(win);
 416
 417	/* without wrapping, first block is the oldest */
 418	if (!msc_block_wrapped(bdesc))
 419		return msc_win_base_sg(win);
 420
 421	/*
 422	 * with wrapping, last written block contains both the newest and the
 423	 * oldest data for this window.
 424	 */
 425	for_each_sg(win->sgt->sgl, sg, win->nr_segs, blk) {
 426		struct msc_block_desc *bdesc = sg_virt(sg);
 427
 428		if (msc_block_last_written(bdesc))
 429			return sg;
 430	}
 431
 432	return msc_win_base_sg(win);
 433}
 434
 435static struct msc_block_desc *msc_iter_bdesc(struct msc_iter *iter)
 436{
 437	return sg_virt(iter->block);
 438}
 439
 440static struct msc_iter *msc_iter_install(struct msc *msc)
 441{
 442	struct msc_iter *iter;
 443
 444	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
 445	if (!iter)
 446		return ERR_PTR(-ENOMEM);
 447
 448	mutex_lock(&msc->buf_mutex);
 449
 450	/*
 451	 * Reading and tracing are mutually exclusive; if msc is
 452	 * enabled, open() will fail; otherwise existing readers
 453	 * will prevent enabling the msc and the rest of fops don't
 454	 * need to worry about it.
 455	 */
 456	if (msc->enabled) {
 457		kfree(iter);
 458		iter = ERR_PTR(-EBUSY);
 459		goto unlock;
 460	}
 461
 462	iter->msc = msc;
 463
 464	list_add_tail(&iter->entry, &msc->iter_list);
 465unlock:
 466	mutex_unlock(&msc->buf_mutex);
 467
 468	return iter;
 469}
 470
 471static void msc_iter_remove(struct msc_iter *iter, struct msc *msc)
 472{
 473	mutex_lock(&msc->buf_mutex);
 474	list_del(&iter->entry);
 475	mutex_unlock(&msc->buf_mutex);
 476
 477	kfree(iter);
 478}
 479
 480static void msc_iter_block_start(struct msc_iter *iter)
 481{
 482	if (iter->start_block)
 483		return;
 484
 485	iter->start_block = msc_win_oldest_sg(iter->win);
 486	iter->block = iter->start_block;
 487	iter->wrap_count = 0;
 488
 489	/*
 490	 * start with the block with oldest data; if data has wrapped
 491	 * in this window, it should be in this block
 492	 */
 493	if (msc_block_wrapped(msc_iter_bdesc(iter)))
 494		iter->wrap_count = 2;
 495
 496}
 497
 498static int msc_iter_win_start(struct msc_iter *iter, struct msc *msc)
 499{
 500	/* already started, nothing to do */
 501	if (iter->start_win)
 502		return 0;
 503
 504	iter->start_win = msc_oldest_window(msc);
 505	if (!iter->start_win)
 506		return -EINVAL;
 507
 508	iter->win = iter->start_win;
 509	iter->start_block = NULL;
 510
 511	msc_iter_block_start(iter);
 512
 513	return 0;
 514}
 515
 516static int msc_iter_win_advance(struct msc_iter *iter)
 517{
 518	iter->win = msc_next_window(iter->win);
 519	iter->start_block = NULL;
 520
 521	if (iter->win == iter->start_win) {
 522		iter->eof++;
 523		return 1;
 524	}
 525
 526	msc_iter_block_start(iter);
 527
 528	return 0;
 529}
 530
 531static int msc_iter_block_advance(struct msc_iter *iter)
 532{
 533	iter->block_off = 0;
 534
 535	/* wrapping */
 536	if (iter->wrap_count && iter->block == iter->start_block) {
 537		iter->wrap_count--;
 538		if (!iter->wrap_count)
 539			/* copied newest data from the wrapped block */
 540			return msc_iter_win_advance(iter);
 541	}
 542
 543	/* no wrapping, check for last written block */
 544	if (!iter->wrap_count && msc_block_last_written(msc_iter_bdesc(iter)))
 545		/* copied newest data for the window */
 546		return msc_iter_win_advance(iter);
 547
 548	/* block advance */
 549	if (sg_is_last(iter->block))
 550		iter->block = msc_win_base_sg(iter->win);
 551	else
 552		iter->block = sg_next(iter->block);
 553
 554	/* no wrapping, sanity check in case there is no last written block */
 555	if (!iter->wrap_count && iter->block == iter->start_block)
 556		return msc_iter_win_advance(iter);
 557
 558	return 0;
 559}
 560
 561/**
 562 * msc_buffer_iterate() - go through multiblock buffer's data
 563 * @iter:	iterator structure
 564 * @size:	amount of data to scan
 565 * @data:	callback's private data
 566 * @fn:		iterator callback
 567 *
 568 * This will start at the window which will be written to next (containing
 569 * the oldest data) and work its way to the current window, calling @fn
 570 * for each chunk of data as it goes.
 571 *
 572 * Caller should have msc::user_count reference to make sure the buffer
 573 * doesn't disappear from under us.
 574 *
 575 * Return:	amount of data actually scanned.
 576 */
 577static ssize_t
 578msc_buffer_iterate(struct msc_iter *iter, size_t size, void *data,
 579		   unsigned long (*fn)(void *, void *, size_t))
 580{
 581	struct msc *msc = iter->msc;
 582	size_t len = size;
 583	unsigned int advance;
 584
 585	if (iter->eof)
 586		return 0;
 587
 588	/* start with the oldest window */
 589	if (msc_iter_win_start(iter, msc))
 590		return 0;
 591
 592	do {
 593		unsigned long data_bytes = msc_data_sz(msc_iter_bdesc(iter));
 594		void *src = (void *)msc_iter_bdesc(iter) + MSC_BDESC;
 595		size_t tocopy = data_bytes, copied = 0;
 596		size_t remaining = 0;
 597
 598		advance = 1;
 599
 600		/*
 601		 * If block wrapping happened, we need to visit the last block
 602		 * twice, because it contains both the oldest and the newest
 603		 * data in this window.
 604		 *
 605		 * First time (wrap_count==2), in the very beginning, to collect
 606		 * the oldest data, which is in the range
 607		 * (data_bytes..DATA_IN_PAGE).
 608		 *
 609		 * Second time (wrap_count==1), it's just like any other block,
 610		 * containing data in the range of [MSC_BDESC..data_bytes].
 611		 */
 612		if (iter->block == iter->start_block && iter->wrap_count == 2) {
 613			tocopy = DATA_IN_PAGE - data_bytes;
 614			src += data_bytes;
 615		}
 616
 617		if (!tocopy)
 618			goto next_block;
 619
 620		tocopy -= iter->block_off;
 621		src += iter->block_off;
 622
 623		if (len < tocopy) {
 624			tocopy = len;
 625			advance = 0;
 626		}
 627
 628		remaining = fn(data, src, tocopy);
 629
 630		if (remaining)
 631			advance = 0;
 632
 633		copied = tocopy - remaining;
 634		len -= copied;
 635		iter->block_off += copied;
 636		iter->offset += copied;
 637
 638		if (!advance)
 639			break;
 640
 641next_block:
 642		if (msc_iter_block_advance(iter))
 643			break;
 644
 645	} while (len);
 646
 647	return size - len;
 648}
 649
 650/**
 651 * msc_buffer_clear_hw_header() - clear hw header for multiblock
 652 * @msc:	MSC device
 653 */
 654static void msc_buffer_clear_hw_header(struct msc *msc)
 655{
 656	struct msc_window *win;
 657	struct scatterlist *sg;
 658
 659	list_for_each_entry(win, &msc->win_list, entry) {
 660		unsigned int blk;
 661		size_t hw_sz = sizeof(struct msc_block_desc) -
 662			offsetof(struct msc_block_desc, hw_tag);
 663
 664		for_each_sg(win->sgt->sgl, sg, win->nr_segs, blk) {
 665			struct msc_block_desc *bdesc = sg_virt(sg);
 666
 667			memset(&bdesc->hw_tag, 0, hw_sz);
 668		}
 669	}
 670}
 671
 672static int intel_th_msu_init(struct msc *msc)
 673{
 674	u32 mintctl, msusts;
 675
 676	if (!msc->do_irq)
 677		return 0;
 678
 679	if (!msc->mbuf)
 680		return 0;
 681
 682	mintctl = ioread32(msc->msu_base + REG_MSU_MINTCTL);
 683	mintctl |= msc->index ? M1BLIE : M0BLIE;
 684	iowrite32(mintctl, msc->msu_base + REG_MSU_MINTCTL);
 685	if (mintctl != ioread32(msc->msu_base + REG_MSU_MINTCTL)) {
 686		dev_info(msc_dev(msc), "MINTCTL ignores writes: no usable interrupts\n");
 687		msc->do_irq = 0;
 688		return 0;
 689	}
 690
 691	msusts = ioread32(msc->msu_base + REG_MSU_MSUSTS);
 692	iowrite32(msusts, msc->msu_base + REG_MSU_MSUSTS);
 693
 694	return 0;
 695}
 696
 697static void intel_th_msu_deinit(struct msc *msc)
 698{
 699	u32 mintctl;
 700
 701	if (!msc->do_irq)
 702		return;
 703
 704	mintctl = ioread32(msc->msu_base + REG_MSU_MINTCTL);
 705	mintctl &= msc->index ? ~M1BLIE : ~M0BLIE;
 706	iowrite32(mintctl, msc->msu_base + REG_MSU_MINTCTL);
 707}
 708
 709static int msc_win_set_lockout(struct msc_window *win,
 710			       enum lockout_state expect,
 711			       enum lockout_state new)
 712{
 713	enum lockout_state old;
 714	unsigned long flags;
 715	int ret = 0;
 716
 717	if (!win->msc->mbuf)
 718		return 0;
 719
 720	spin_lock_irqsave(&win->lo_lock, flags);
 721	old = win->lockout;
 722
 723	if (old != expect) {
 724		ret = -EINVAL;
 
 
 
 725		goto unlock;
 726	}
 727
 728	win->lockout = new;
 729
 730	if (old == expect && new == WIN_LOCKED)
 731		atomic_inc(&win->msc->user_count);
 732	else if (old == expect && old == WIN_LOCKED)
 733		atomic_dec(&win->msc->user_count);
 734
 735unlock:
 736	spin_unlock_irqrestore(&win->lo_lock, flags);
 737
 738	if (ret) {
 739		if (expect == WIN_READY && old == WIN_LOCKED)
 740			return -EBUSY;
 741
 742		/* from intel_th_msc_window_unlock(), don't warn if not locked */
 743		if (expect == WIN_LOCKED && old == new)
 744			return 0;
 745
 746		dev_warn_ratelimited(msc_dev(win->msc),
 747				     "expected lockout state %d, got %d\n",
 748				     expect, old);
 749	}
 750
 751	return ret;
 752}
 753/**
 754 * msc_configure() - set up MSC hardware
 755 * @msc:	the MSC device to configure
 756 *
 757 * Program storage mode, wrapping, burst length and trace buffer address
 758 * into a given MSC. Then, enable tracing and set msc::enabled.
 759 * The latter is serialized on msc::buf_mutex, so make sure to hold it.
 760 */
 761static int msc_configure(struct msc *msc)
 762{
 763	u32 reg;
 764
 765	lockdep_assert_held(&msc->buf_mutex);
 766
 767	if (msc->mode > MSC_MODE_MULTI)
 768		return -EINVAL;
 769
 770	if (msc->mode == MSC_MODE_MULTI) {
 771		if (msc_win_set_lockout(msc->cur_win, WIN_READY, WIN_INUSE))
 772			return -EBUSY;
 773
 774		msc_buffer_clear_hw_header(msc);
 775	}
 776
 777	msc->orig_addr = ioread32(msc->reg_base + REG_MSU_MSC0BAR);
 778	msc->orig_sz   = ioread32(msc->reg_base + REG_MSU_MSC0SIZE);
 779
 780	reg = msc->base_addr >> PAGE_SHIFT;
 781	iowrite32(reg, msc->reg_base + REG_MSU_MSC0BAR);
 782
 783	if (msc->mode == MSC_MODE_SINGLE) {
 784		reg = msc->nr_pages;
 785		iowrite32(reg, msc->reg_base + REG_MSU_MSC0SIZE);
 786	}
 787
 788	reg = ioread32(msc->reg_base + REG_MSU_MSC0CTL);
 789	reg &= ~(MSC_MODE | MSC_WRAPEN | MSC_EN | MSC_RD_HDR_OVRD);
 790
 791	reg |= MSC_EN;
 792	reg |= msc->mode << __ffs(MSC_MODE);
 793	reg |= msc->burst_len << __ffs(MSC_LEN);
 794
 795	if (msc->wrap)
 796		reg |= MSC_WRAPEN;
 797
 798	iowrite32(reg, msc->reg_base + REG_MSU_MSC0CTL);
 799
 800	intel_th_msu_init(msc);
 801
 802	msc->thdev->output.multiblock = msc->mode == MSC_MODE_MULTI;
 803	intel_th_trace_enable(msc->thdev);
 804	msc->enabled = 1;
 805
 806	if (msc->mbuf && msc->mbuf->activate)
 807		msc->mbuf->activate(msc->mbuf_priv);
 808
 809	return 0;
 810}
 811
 812/**
 813 * msc_disable() - disable MSC hardware
 814 * @msc:	MSC device to disable
 815 *
 816 * If @msc is enabled, disable tracing on the switch and then disable MSC
 817 * storage. Caller must hold msc::buf_mutex.
 818 */
 819static void msc_disable(struct msc *msc)
 820{
 821	struct msc_window *win = msc->cur_win;
 822	u32 reg;
 823
 824	lockdep_assert_held(&msc->buf_mutex);
 825
 826	if (msc->mode == MSC_MODE_MULTI)
 827		msc_win_set_lockout(win, WIN_INUSE, WIN_LOCKED);
 828
 829	if (msc->mbuf && msc->mbuf->deactivate)
 830		msc->mbuf->deactivate(msc->mbuf_priv);
 831	intel_th_msu_deinit(msc);
 832	intel_th_trace_disable(msc->thdev);
 833
 834	if (msc->mode == MSC_MODE_SINGLE) {
 835		reg = ioread32(msc->reg_base + REG_MSU_MSC0STS);
 836		msc->single_wrap = !!(reg & MSCSTS_WRAPSTAT);
 837
 838		reg = ioread32(msc->reg_base + REG_MSU_MSC0MWP);
 839		msc->single_sz = reg & ((msc->nr_pages << PAGE_SHIFT) - 1);
 840		dev_dbg(msc_dev(msc), "MSCnMWP: %08x/%08lx, wrap: %d\n",
 841			reg, msc->single_sz, msc->single_wrap);
 842	}
 843
 844	reg = ioread32(msc->reg_base + REG_MSU_MSC0CTL);
 845	reg &= ~MSC_EN;
 846	iowrite32(reg, msc->reg_base + REG_MSU_MSC0CTL);
 847
 848	if (msc->mbuf && msc->mbuf->ready)
 849		msc->mbuf->ready(msc->mbuf_priv, win->sgt,
 850				 msc_win_total_sz(win));
 851
 852	msc->enabled = 0;
 853
 854	iowrite32(msc->orig_addr, msc->reg_base + REG_MSU_MSC0BAR);
 855	iowrite32(msc->orig_sz, msc->reg_base + REG_MSU_MSC0SIZE);
 856
 857	dev_dbg(msc_dev(msc), "MSCnNWSA: %08x\n",
 858		ioread32(msc->reg_base + REG_MSU_MSC0NWSA));
 859
 860	reg = ioread32(msc->reg_base + REG_MSU_MSC0STS);
 861	dev_dbg(msc_dev(msc), "MSCnSTS: %08x\n", reg);
 862
 863	reg = ioread32(msc->reg_base + REG_MSU_MSUSTS);
 864	reg &= msc->index ? MSUSTS_MSC1BLAST : MSUSTS_MSC0BLAST;
 865	iowrite32(reg, msc->reg_base + REG_MSU_MSUSTS);
 866}
 867
 868static int intel_th_msc_activate(struct intel_th_device *thdev)
 869{
 870	struct msc *msc = dev_get_drvdata(&thdev->dev);
 871	int ret = -EBUSY;
 872
 873	if (!atomic_inc_unless_negative(&msc->user_count))
 874		return -ENODEV;
 875
 876	mutex_lock(&msc->buf_mutex);
 877
 878	/* if there are readers, refuse */
 879	if (list_empty(&msc->iter_list))
 880		ret = msc_configure(msc);
 881
 882	mutex_unlock(&msc->buf_mutex);
 883
 884	if (ret)
 885		atomic_dec(&msc->user_count);
 886
 887	return ret;
 888}
 889
 890static void intel_th_msc_deactivate(struct intel_th_device *thdev)
 891{
 892	struct msc *msc = dev_get_drvdata(&thdev->dev);
 893
 894	mutex_lock(&msc->buf_mutex);
 895	if (msc->enabled) {
 896		msc_disable(msc);
 897		atomic_dec(&msc->user_count);
 898	}
 899	mutex_unlock(&msc->buf_mutex);
 900}
 901
 902/**
 903 * msc_buffer_contig_alloc() - allocate a contiguous buffer for SINGLE mode
 904 * @msc:	MSC device
 905 * @size:	allocation size in bytes
 906 *
 907 * This modifies msc::base, which requires msc::buf_mutex to serialize, so the
 908 * caller is expected to hold it.
 909 *
 910 * Return:	0 on success, -errno otherwise.
 911 */
 912static int msc_buffer_contig_alloc(struct msc *msc, unsigned long size)
 913{
 914	unsigned long nr_pages = size >> PAGE_SHIFT;
 915	unsigned int order = get_order(size);
 916	struct page *page;
 917	int ret;
 918
 919	if (!size)
 920		return 0;
 921
 922	ret = sg_alloc_table(&msc->single_sgt, 1, GFP_KERNEL);
 923	if (ret)
 924		goto err_out;
 925
 926	ret = -ENOMEM;
 927	page = alloc_pages(GFP_KERNEL | __GFP_ZERO | GFP_DMA32, order);
 928	if (!page)
 929		goto err_free_sgt;
 930
 931	split_page(page, order);
 932	sg_set_buf(msc->single_sgt.sgl, page_address(page), size);
 933
 934	ret = dma_map_sg(msc_dev(msc)->parent->parent, msc->single_sgt.sgl, 1,
 935			 DMA_FROM_DEVICE);
 936	if (ret < 0)
 937		goto err_free_pages;
 938
 939	msc->nr_pages = nr_pages;
 940	msc->base = page_address(page);
 941	msc->base_addr = sg_dma_address(msc->single_sgt.sgl);
 942
 943	return 0;
 944
 945err_free_pages:
 946	__free_pages(page, order);
 947
 948err_free_sgt:
 949	sg_free_table(&msc->single_sgt);
 950
 951err_out:
 952	return ret;
 953}
 954
 955/**
 956 * msc_buffer_contig_free() - free a contiguous buffer
 957 * @msc:	MSC configured in SINGLE mode
 958 */
 959static void msc_buffer_contig_free(struct msc *msc)
 960{
 961	unsigned long off;
 962
 963	dma_unmap_sg(msc_dev(msc)->parent->parent, msc->single_sgt.sgl,
 964		     1, DMA_FROM_DEVICE);
 965	sg_free_table(&msc->single_sgt);
 966
 967	for (off = 0; off < msc->nr_pages << PAGE_SHIFT; off += PAGE_SIZE) {
 968		struct page *page = virt_to_page(msc->base + off);
 969
 970		page->mapping = NULL;
 971		__free_page(page);
 972	}
 973
 974	msc->nr_pages = 0;
 975}
 976
 977/**
 978 * msc_buffer_contig_get_page() - find a page at a given offset
 979 * @msc:	MSC configured in SINGLE mode
 980 * @pgoff:	page offset
 981 *
 982 * Return:	page, if @pgoff is within the range, NULL otherwise.
 983 */
 984static struct page *msc_buffer_contig_get_page(struct msc *msc,
 985					       unsigned long pgoff)
 986{
 987	if (pgoff >= msc->nr_pages)
 988		return NULL;
 989
 990	return virt_to_page(msc->base + (pgoff << PAGE_SHIFT));
 991}
 992
 993static int __msc_buffer_win_alloc(struct msc_window *win,
 994				  unsigned int nr_segs)
 995{
 996	struct scatterlist *sg_ptr;
 997	void *block;
 998	int i, ret;
 999
1000	ret = sg_alloc_table(win->sgt, nr_segs, GFP_KERNEL);
1001	if (ret)
1002		return -ENOMEM;
1003
1004	for_each_sg(win->sgt->sgl, sg_ptr, nr_segs, i) {
1005		block = dma_alloc_coherent(msc_dev(win->msc)->parent->parent,
1006					  PAGE_SIZE, &sg_dma_address(sg_ptr),
1007					  GFP_KERNEL);
1008		if (!block)
1009			goto err_nomem;
1010
1011		sg_set_buf(sg_ptr, block, PAGE_SIZE);
1012	}
1013
1014	return nr_segs;
1015
1016err_nomem:
1017	for_each_sg(win->sgt->sgl, sg_ptr, i, ret)
1018		dma_free_coherent(msc_dev(win->msc)->parent->parent, PAGE_SIZE,
1019				  sg_virt(sg_ptr), sg_dma_address(sg_ptr));
1020
1021	sg_free_table(win->sgt);
1022
1023	return -ENOMEM;
1024}
1025
1026#ifdef CONFIG_X86
1027static void msc_buffer_set_uc(struct msc_window *win, unsigned int nr_segs)
1028{
1029	struct scatterlist *sg_ptr;
1030	int i;
1031
1032	for_each_sg(win->sgt->sgl, sg_ptr, nr_segs, i) {
1033		/* Set the page as uncached */
1034		set_memory_uc((unsigned long)sg_virt(sg_ptr),
1035			      PFN_DOWN(sg_ptr->length));
1036	}
1037}
1038
1039static void msc_buffer_set_wb(struct msc_window *win)
1040{
1041	struct scatterlist *sg_ptr;
1042	int i;
1043
1044	for_each_sg(win->sgt->sgl, sg_ptr, win->nr_segs, i) {
1045		/* Reset the page to write-back */
1046		set_memory_wb((unsigned long)sg_virt(sg_ptr),
1047			      PFN_DOWN(sg_ptr->length));
1048	}
1049}
1050#else /* !X86 */
1051static inline void
1052msc_buffer_set_uc(struct msc_window *win, unsigned int nr_segs) {}
1053static inline void msc_buffer_set_wb(struct msc_window *win) {}
1054#endif /* CONFIG_X86 */
1055
1056/**
1057 * msc_buffer_win_alloc() - alloc a window for a multiblock mode
1058 * @msc:	MSC device
1059 * @nr_blocks:	number of pages in this window
1060 *
1061 * This modifies msc::win_list and msc::base, which requires msc::buf_mutex
1062 * to serialize, so the caller is expected to hold it.
1063 *
1064 * Return:	0 on success, -errno otherwise.
1065 */
1066static int msc_buffer_win_alloc(struct msc *msc, unsigned int nr_blocks)
1067{
1068	struct msc_window *win;
1069	int ret = -ENOMEM;
1070
1071	if (!nr_blocks)
1072		return 0;
1073
1074	win = kzalloc(sizeof(*win), GFP_KERNEL);
1075	if (!win)
1076		return -ENOMEM;
1077
1078	win->msc = msc;
1079	win->sgt = &win->_sgt;
1080	win->lockout = WIN_READY;
1081	spin_lock_init(&win->lo_lock);
1082
1083	if (!list_empty(&msc->win_list)) {
1084		struct msc_window *prev = list_last_entry(&msc->win_list,
1085							  struct msc_window,
1086							  entry);
1087
1088		win->pgoff = prev->pgoff + prev->nr_blocks;
1089	}
1090
1091	if (msc->mbuf && msc->mbuf->alloc_window)
1092		ret = msc->mbuf->alloc_window(msc->mbuf_priv, &win->sgt,
1093					      nr_blocks << PAGE_SHIFT);
1094	else
1095		ret = __msc_buffer_win_alloc(win, nr_blocks);
1096
1097	if (ret <= 0)
1098		goto err_nomem;
1099
1100	msc_buffer_set_uc(win, ret);
1101
1102	win->nr_segs = ret;
1103	win->nr_blocks = nr_blocks;
1104
1105	if (list_empty(&msc->win_list)) {
1106		msc->base = msc_win_base(win);
1107		msc->base_addr = msc_win_base_dma(win);
1108		msc->cur_win = win;
1109	}
1110
1111	list_add_tail(&win->entry, &msc->win_list);
1112	msc->nr_pages += nr_blocks;
1113
1114	return 0;
1115
1116err_nomem:
1117	kfree(win);
1118
1119	return ret;
1120}
1121
1122static void __msc_buffer_win_free(struct msc *msc, struct msc_window *win)
1123{
1124	struct scatterlist *sg;
1125	int i;
1126
1127	for_each_sg(win->sgt->sgl, sg, win->nr_segs, i) {
1128		struct page *page = sg_page(sg);
1129
1130		page->mapping = NULL;
1131		dma_free_coherent(msc_dev(win->msc)->parent->parent, PAGE_SIZE,
1132				  sg_virt(sg), sg_dma_address(sg));
1133	}
1134	sg_free_table(win->sgt);
1135}
1136
1137/**
1138 * msc_buffer_win_free() - free a window from MSC's window list
1139 * @msc:	MSC device
1140 * @win:	window to free
1141 *
1142 * This modifies msc::win_list and msc::base, which requires msc::buf_mutex
1143 * to serialize, so the caller is expected to hold it.
1144 */
1145static void msc_buffer_win_free(struct msc *msc, struct msc_window *win)
1146{
1147	msc->nr_pages -= win->nr_blocks;
1148
1149	list_del(&win->entry);
1150	if (list_empty(&msc->win_list)) {
1151		msc->base = NULL;
1152		msc->base_addr = 0;
1153	}
1154
1155	msc_buffer_set_wb(win);
1156
1157	if (msc->mbuf && msc->mbuf->free_window)
1158		msc->mbuf->free_window(msc->mbuf_priv, win->sgt);
1159	else
1160		__msc_buffer_win_free(msc, win);
1161
1162	kfree(win);
1163}
1164
1165/**
1166 * msc_buffer_relink() - set up block descriptors for multiblock mode
1167 * @msc:	MSC device
1168 *
1169 * This traverses msc::win_list, which requires msc::buf_mutex to serialize,
1170 * so the caller is expected to hold it.
1171 */
1172static void msc_buffer_relink(struct msc *msc)
1173{
1174	struct msc_window *win, *next_win;
1175
1176	/* call with msc::mutex locked */
1177	list_for_each_entry(win, &msc->win_list, entry) {
1178		struct scatterlist *sg;
1179		unsigned int blk;
1180		u32 sw_tag = 0;
1181
1182		/*
1183		 * Last window's next_win should point to the first window
1184		 * and MSC_SW_TAG_LASTWIN should be set.
1185		 */
1186		if (msc_is_last_win(win)) {
1187			sw_tag |= MSC_SW_TAG_LASTWIN;
1188			next_win = list_first_entry(&msc->win_list,
1189						    struct msc_window, entry);
1190		} else {
1191			next_win = list_next_entry(win, entry);
1192		}
1193
1194		for_each_sg(win->sgt->sgl, sg, win->nr_segs, blk) {
1195			struct msc_block_desc *bdesc = sg_virt(sg);
1196
1197			memset(bdesc, 0, sizeof(*bdesc));
1198
1199			bdesc->next_win = msc_win_base_pfn(next_win);
1200
1201			/*
1202			 * Similarly to last window, last block should point
1203			 * to the first one.
1204			 */
1205			if (blk == win->nr_segs - 1) {
1206				sw_tag |= MSC_SW_TAG_LASTBLK;
1207				bdesc->next_blk = msc_win_base_pfn(win);
1208			} else {
1209				dma_addr_t addr = sg_dma_address(sg_next(sg));
1210
1211				bdesc->next_blk = PFN_DOWN(addr);
1212			}
1213
1214			bdesc->sw_tag = sw_tag;
1215			bdesc->block_sz = sg->length / 64;
1216		}
1217	}
1218
1219	/*
1220	 * Make the above writes globally visible before tracing is
1221	 * enabled to make sure hardware sees them coherently.
1222	 */
1223	wmb();
1224}
1225
1226static void msc_buffer_multi_free(struct msc *msc)
1227{
1228	struct msc_window *win, *iter;
1229
1230	list_for_each_entry_safe(win, iter, &msc->win_list, entry)
1231		msc_buffer_win_free(msc, win);
1232}
1233
1234static int msc_buffer_multi_alloc(struct msc *msc, unsigned long *nr_pages,
1235				  unsigned int nr_wins)
1236{
1237	int ret, i;
1238
1239	for (i = 0; i < nr_wins; i++) {
1240		ret = msc_buffer_win_alloc(msc, nr_pages[i]);
1241		if (ret) {
1242			msc_buffer_multi_free(msc);
1243			return ret;
1244		}
1245	}
1246
1247	msc_buffer_relink(msc);
1248
1249	return 0;
1250}
1251
1252/**
1253 * msc_buffer_free() - free buffers for MSC
1254 * @msc:	MSC device
1255 *
1256 * Free MSC's storage buffers.
1257 *
1258 * This modifies msc::win_list and msc::base, which requires msc::buf_mutex to
1259 * serialize, so the caller is expected to hold it.
1260 */
1261static void msc_buffer_free(struct msc *msc)
1262{
1263	if (msc->mode == MSC_MODE_SINGLE)
1264		msc_buffer_contig_free(msc);
1265	else if (msc->mode == MSC_MODE_MULTI)
1266		msc_buffer_multi_free(msc);
1267}
1268
1269/**
1270 * msc_buffer_alloc() - allocate a buffer for MSC
1271 * @msc:	MSC device
1272 * @size:	allocation size in bytes
1273 *
1274 * Allocate a storage buffer for MSC, depending on the msc::mode, it will be
1275 * either done via msc_buffer_contig_alloc() for SINGLE operation mode or
1276 * msc_buffer_win_alloc() for multiblock operation. The latter allocates one
1277 * window per invocation, so in multiblock mode this can be called multiple
1278 * times for the same MSC to allocate multiple windows.
1279 *
1280 * This modifies msc::win_list and msc::base, which requires msc::buf_mutex
1281 * to serialize, so the caller is expected to hold it.
1282 *
1283 * Return:	0 on success, -errno otherwise.
1284 */
1285static int msc_buffer_alloc(struct msc *msc, unsigned long *nr_pages,
1286			    unsigned int nr_wins)
1287{
1288	int ret;
1289
1290	/* -1: buffer not allocated */
1291	if (atomic_read(&msc->user_count) != -1)
1292		return -EBUSY;
1293
1294	if (msc->mode == MSC_MODE_SINGLE) {
1295		if (nr_wins != 1)
1296			return -EINVAL;
1297
1298		ret = msc_buffer_contig_alloc(msc, nr_pages[0] << PAGE_SHIFT);
1299	} else if (msc->mode == MSC_MODE_MULTI) {
1300		ret = msc_buffer_multi_alloc(msc, nr_pages, nr_wins);
1301	} else {
1302		ret = -EINVAL;
1303	}
1304
1305	if (!ret) {
1306		/* allocation should be visible before the counter goes to 0 */
1307		smp_mb__before_atomic();
1308
1309		if (WARN_ON_ONCE(atomic_cmpxchg(&msc->user_count, -1, 0) != -1))
1310			return -EINVAL;
1311	}
1312
1313	return ret;
1314}
1315
1316/**
1317 * msc_buffer_unlocked_free_unless_used() - free a buffer unless it's in use
1318 * @msc:	MSC device
1319 *
1320 * This will free MSC buffer unless it is in use or there is no allocated
1321 * buffer.
1322 * Caller needs to hold msc::buf_mutex.
1323 *
1324 * Return:	0 on successful deallocation or if there was no buffer to
1325 *		deallocate, -EBUSY if there are active users.
1326 */
1327static int msc_buffer_unlocked_free_unless_used(struct msc *msc)
1328{
1329	int count, ret = 0;
1330
1331	count = atomic_cmpxchg(&msc->user_count, 0, -1);
1332
1333	/* > 0: buffer is allocated and has users */
1334	if (count > 0)
1335		ret = -EBUSY;
1336	/* 0: buffer is allocated, no users */
1337	else if (!count)
1338		msc_buffer_free(msc);
1339	/* < 0: no buffer, nothing to do */
1340
1341	return ret;
1342}
1343
1344/**
1345 * msc_buffer_free_unless_used() - free a buffer unless it's in use
1346 * @msc:	MSC device
1347 *
1348 * This is a locked version of msc_buffer_unlocked_free_unless_used().
1349 */
1350static int msc_buffer_free_unless_used(struct msc *msc)
1351{
1352	int ret;
1353
1354	mutex_lock(&msc->buf_mutex);
1355	ret = msc_buffer_unlocked_free_unless_used(msc);
1356	mutex_unlock(&msc->buf_mutex);
1357
1358	return ret;
1359}
1360
1361/**
1362 * msc_buffer_get_page() - get MSC buffer page at a given offset
1363 * @msc:	MSC device
1364 * @pgoff:	page offset into the storage buffer
1365 *
1366 * This traverses msc::win_list, so holding msc::buf_mutex is expected from
1367 * the caller.
1368 *
1369 * Return:	page if @pgoff corresponds to a valid buffer page or NULL.
1370 */
1371static struct page *msc_buffer_get_page(struct msc *msc, unsigned long pgoff)
1372{
1373	struct msc_window *win;
1374	struct scatterlist *sg;
1375	unsigned int blk;
1376
1377	if (msc->mode == MSC_MODE_SINGLE)
1378		return msc_buffer_contig_get_page(msc, pgoff);
1379
1380	list_for_each_entry(win, &msc->win_list, entry)
1381		if (pgoff >= win->pgoff && pgoff < win->pgoff + win->nr_blocks)
1382			goto found;
1383
1384	return NULL;
1385
1386found:
1387	pgoff -= win->pgoff;
1388
1389	for_each_sg(win->sgt->sgl, sg, win->nr_segs, blk) {
1390		struct page *page = sg_page(sg);
1391		size_t pgsz = PFN_DOWN(sg->length);
1392
1393		if (pgoff < pgsz)
1394			return page + pgoff;
1395
1396		pgoff -= pgsz;
1397	}
1398
1399	return NULL;
1400}
1401
1402/**
1403 * struct msc_win_to_user_struct - data for copy_to_user() callback
1404 * @buf:	userspace buffer to copy data to
1405 * @offset:	running offset
1406 */
1407struct msc_win_to_user_struct {
1408	char __user	*buf;
1409	unsigned long	offset;
1410};
1411
1412/**
1413 * msc_win_to_user() - iterator for msc_buffer_iterate() to copy data to user
1414 * @data:	callback's private data
1415 * @src:	source buffer
1416 * @len:	amount of data to copy from the source buffer
1417 */
1418static unsigned long msc_win_to_user(void *data, void *src, size_t len)
1419{
1420	struct msc_win_to_user_struct *u = data;
1421	unsigned long ret;
1422
1423	ret = copy_to_user(u->buf + u->offset, src, len);
1424	u->offset += len - ret;
1425
1426	return ret;
1427}
1428
1429
1430/*
1431 * file operations' callbacks
1432 */
1433
1434static int intel_th_msc_open(struct inode *inode, struct file *file)
1435{
1436	struct intel_th_device *thdev = file->private_data;
1437	struct msc *msc = dev_get_drvdata(&thdev->dev);
1438	struct msc_iter *iter;
1439
1440	if (!capable(CAP_SYS_RAWIO))
1441		return -EPERM;
1442
1443	iter = msc_iter_install(msc);
1444	if (IS_ERR(iter))
1445		return PTR_ERR(iter);
1446
1447	file->private_data = iter;
1448
1449	return nonseekable_open(inode, file);
1450}
1451
1452static int intel_th_msc_release(struct inode *inode, struct file *file)
1453{
1454	struct msc_iter *iter = file->private_data;
1455	struct msc *msc = iter->msc;
1456
1457	msc_iter_remove(iter, msc);
1458
1459	return 0;
1460}
1461
1462static ssize_t
1463msc_single_to_user(struct msc *msc, char __user *buf, loff_t off, size_t len)
1464{
1465	unsigned long size = msc->nr_pages << PAGE_SHIFT, rem = len;
1466	unsigned long start = off, tocopy = 0;
1467
1468	if (msc->single_wrap) {
1469		start += msc->single_sz;
1470		if (start < size) {
1471			tocopy = min(rem, size - start);
1472			if (copy_to_user(buf, msc->base + start, tocopy))
1473				return -EFAULT;
1474
1475			buf += tocopy;
1476			rem -= tocopy;
1477			start += tocopy;
1478		}
1479
1480		start &= size - 1;
1481		if (rem) {
1482			tocopy = min(rem, msc->single_sz - start);
1483			if (copy_to_user(buf, msc->base + start, tocopy))
1484				return -EFAULT;
1485
1486			rem -= tocopy;
1487		}
1488
1489		return len - rem;
1490	}
1491
1492	if (copy_to_user(buf, msc->base + start, rem))
1493		return -EFAULT;
1494
1495	return len;
1496}
1497
1498static ssize_t intel_th_msc_read(struct file *file, char __user *buf,
1499				 size_t len, loff_t *ppos)
1500{
1501	struct msc_iter *iter = file->private_data;
1502	struct msc *msc = iter->msc;
1503	size_t size;
1504	loff_t off = *ppos;
1505	ssize_t ret = 0;
1506
1507	if (!atomic_inc_unless_negative(&msc->user_count))
1508		return 0;
1509
1510	if (msc->mode == MSC_MODE_SINGLE && !msc->single_wrap)
1511		size = msc->single_sz;
1512	else
1513		size = msc->nr_pages << PAGE_SHIFT;
1514
1515	if (!size)
1516		goto put_count;
1517
1518	if (off >= size)
1519		goto put_count;
1520
1521	if (off + len >= size)
1522		len = size - off;
1523
1524	if (msc->mode == MSC_MODE_SINGLE) {
1525		ret = msc_single_to_user(msc, buf, off, len);
1526		if (ret >= 0)
1527			*ppos += ret;
1528	} else if (msc->mode == MSC_MODE_MULTI) {
1529		struct msc_win_to_user_struct u = {
1530			.buf	= buf,
1531			.offset	= 0,
1532		};
1533
1534		ret = msc_buffer_iterate(iter, len, &u, msc_win_to_user);
1535		if (ret >= 0)
1536			*ppos = iter->offset;
1537	} else {
1538		ret = -EINVAL;
1539	}
1540
1541put_count:
1542	atomic_dec(&msc->user_count);
1543
1544	return ret;
1545}
1546
1547/*
1548 * vm operations callbacks (vm_ops)
1549 */
1550
1551static void msc_mmap_open(struct vm_area_struct *vma)
1552{
1553	struct msc_iter *iter = vma->vm_file->private_data;
1554	struct msc *msc = iter->msc;
1555
1556	atomic_inc(&msc->mmap_count);
1557}
1558
1559static void msc_mmap_close(struct vm_area_struct *vma)
1560{
1561	struct msc_iter *iter = vma->vm_file->private_data;
1562	struct msc *msc = iter->msc;
1563	unsigned long pg;
1564
1565	if (!atomic_dec_and_mutex_lock(&msc->mmap_count, &msc->buf_mutex))
1566		return;
1567
1568	/* drop page _refcounts */
1569	for (pg = 0; pg < msc->nr_pages; pg++) {
1570		struct page *page = msc_buffer_get_page(msc, pg);
1571
1572		if (WARN_ON_ONCE(!page))
1573			continue;
1574
1575		if (page->mapping)
1576			page->mapping = NULL;
1577	}
1578
1579	/* last mapping -- drop user_count */
1580	atomic_dec(&msc->user_count);
1581	mutex_unlock(&msc->buf_mutex);
1582}
1583
1584static vm_fault_t msc_mmap_fault(struct vm_fault *vmf)
1585{
1586	struct msc_iter *iter = vmf->vma->vm_file->private_data;
1587	struct msc *msc = iter->msc;
1588
1589	vmf->page = msc_buffer_get_page(msc, vmf->pgoff);
1590	if (!vmf->page)
1591		return VM_FAULT_SIGBUS;
1592
1593	get_page(vmf->page);
1594	vmf->page->mapping = vmf->vma->vm_file->f_mapping;
1595	vmf->page->index = vmf->pgoff;
1596
1597	return 0;
1598}
1599
1600static const struct vm_operations_struct msc_mmap_ops = {
1601	.open	= msc_mmap_open,
1602	.close	= msc_mmap_close,
1603	.fault	= msc_mmap_fault,
1604};
1605
1606static int intel_th_msc_mmap(struct file *file, struct vm_area_struct *vma)
1607{
1608	unsigned long size = vma->vm_end - vma->vm_start;
1609	struct msc_iter *iter = vma->vm_file->private_data;
1610	struct msc *msc = iter->msc;
1611	int ret = -EINVAL;
1612
1613	if (!size || offset_in_page(size))
1614		return -EINVAL;
1615
1616	if (vma->vm_pgoff)
1617		return -EINVAL;
1618
1619	/* grab user_count once per mmap; drop in msc_mmap_close() */
1620	if (!atomic_inc_unless_negative(&msc->user_count))
1621		return -EINVAL;
1622
1623	if (msc->mode != MSC_MODE_SINGLE &&
1624	    msc->mode != MSC_MODE_MULTI)
1625		goto out;
1626
1627	if (size >> PAGE_SHIFT != msc->nr_pages)
1628		goto out;
1629
1630	atomic_set(&msc->mmap_count, 1);
1631	ret = 0;
1632
1633out:
1634	if (ret)
1635		atomic_dec(&msc->user_count);
1636
1637	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1638	vma->vm_flags |= VM_DONTEXPAND | VM_DONTCOPY;
1639	vma->vm_ops = &msc_mmap_ops;
1640	return ret;
1641}
1642
1643static const struct file_operations intel_th_msc_fops = {
1644	.open		= intel_th_msc_open,
1645	.release	= intel_th_msc_release,
1646	.read		= intel_th_msc_read,
1647	.mmap		= intel_th_msc_mmap,
1648	.llseek		= no_llseek,
1649	.owner		= THIS_MODULE,
1650};
1651
1652static void intel_th_msc_wait_empty(struct intel_th_device *thdev)
1653{
1654	struct msc *msc = dev_get_drvdata(&thdev->dev);
1655	unsigned long count;
1656	u32 reg;
1657
1658	for (reg = 0, count = MSC_PLE_WAITLOOP_DEPTH;
1659	     count && !(reg & MSCSTS_PLE); count--) {
1660		reg = __raw_readl(msc->reg_base + REG_MSU_MSC0STS);
1661		cpu_relax();
1662	}
1663
1664	if (!count)
1665		dev_dbg(msc_dev(msc), "timeout waiting for MSC0 PLE\n");
1666}
1667
1668static int intel_th_msc_init(struct msc *msc)
1669{
1670	atomic_set(&msc->user_count, -1);
1671
1672	msc->mode = msc->multi_is_broken ? MSC_MODE_SINGLE : MSC_MODE_MULTI;
1673	mutex_init(&msc->buf_mutex);
1674	INIT_LIST_HEAD(&msc->win_list);
1675	INIT_LIST_HEAD(&msc->iter_list);
1676
1677	msc->burst_len =
1678		(ioread32(msc->reg_base + REG_MSU_MSC0CTL) & MSC_LEN) >>
1679		__ffs(MSC_LEN);
1680
1681	return 0;
1682}
1683
1684static int msc_win_switch(struct msc *msc)
1685{
1686	struct msc_window *first;
1687
1688	if (list_empty(&msc->win_list))
1689		return -EINVAL;
1690
1691	first = list_first_entry(&msc->win_list, struct msc_window, entry);
1692
1693	if (msc_is_last_win(msc->cur_win))
1694		msc->cur_win = first;
1695	else
1696		msc->cur_win = list_next_entry(msc->cur_win, entry);
1697
1698	msc->base = msc_win_base(msc->cur_win);
1699	msc->base_addr = msc_win_base_dma(msc->cur_win);
1700
1701	intel_th_trace_switch(msc->thdev);
1702
1703	return 0;
1704}
1705
1706/**
1707 * intel_th_msc_window_unlock - put the window back in rotation
1708 * @dev:	MSC device to which this relates
1709 * @sgt:	buffer's sg_table for the window, does nothing if NULL
1710 */
1711void intel_th_msc_window_unlock(struct device *dev, struct sg_table *sgt)
1712{
1713	struct msc *msc = dev_get_drvdata(dev);
1714	struct msc_window *win;
1715
1716	if (!sgt)
1717		return;
1718
1719	win = msc_find_window(msc, sgt, false);
1720	if (!win)
1721		return;
1722
1723	msc_win_set_lockout(win, WIN_LOCKED, WIN_READY);
1724	if (msc->switch_on_unlock == win) {
1725		msc->switch_on_unlock = NULL;
1726		msc_win_switch(msc);
1727	}
1728}
1729EXPORT_SYMBOL_GPL(intel_th_msc_window_unlock);
1730
1731static void msc_work(struct work_struct *work)
1732{
1733	struct msc *msc = container_of(work, struct msc, work);
1734
1735	intel_th_msc_deactivate(msc->thdev);
1736}
1737
1738static irqreturn_t intel_th_msc_interrupt(struct intel_th_device *thdev)
1739{
1740	struct msc *msc = dev_get_drvdata(&thdev->dev);
1741	u32 msusts = ioread32(msc->msu_base + REG_MSU_MSUSTS);
1742	u32 mask = msc->index ? MSUSTS_MSC1BLAST : MSUSTS_MSC0BLAST;
1743	struct msc_window *win, *next_win;
1744
1745	if (!msc->do_irq || !msc->mbuf)
1746		return IRQ_NONE;
1747
1748	msusts &= mask;
1749
1750	if (!msusts)
1751		return msc->enabled ? IRQ_HANDLED : IRQ_NONE;
1752
1753	iowrite32(msusts, msc->msu_base + REG_MSU_MSUSTS);
1754
1755	if (!msc->enabled)
1756		return IRQ_NONE;
1757
1758	/* grab the window before we do the switch */
1759	win = msc->cur_win;
1760	if (!win)
1761		return IRQ_HANDLED;
1762	next_win = msc_next_window(win);
1763	if (!next_win)
1764		return IRQ_HANDLED;
1765
1766	/* next window: if READY, proceed, if LOCKED, stop the trace */
1767	if (msc_win_set_lockout(next_win, WIN_READY, WIN_INUSE)) {
1768		if (msc->stop_on_full)
1769			schedule_work(&msc->work);
1770		else
1771			msc->switch_on_unlock = next_win;
1772
1773		return IRQ_HANDLED;
1774	}
1775
1776	/* current window: INUSE -> LOCKED */
1777	msc_win_set_lockout(win, WIN_INUSE, WIN_LOCKED);
1778
1779	msc_win_switch(msc);
1780
1781	if (msc->mbuf && msc->mbuf->ready)
1782		msc->mbuf->ready(msc->mbuf_priv, win->sgt,
1783				 msc_win_total_sz(win));
1784
1785	return IRQ_HANDLED;
1786}
1787
1788static const char * const msc_mode[] = {
1789	[MSC_MODE_SINGLE]	= "single",
1790	[MSC_MODE_MULTI]	= "multi",
1791	[MSC_MODE_EXI]		= "ExI",
1792	[MSC_MODE_DEBUG]	= "debug",
1793};
1794
1795static ssize_t
1796wrap_show(struct device *dev, struct device_attribute *attr, char *buf)
1797{
1798	struct msc *msc = dev_get_drvdata(dev);
1799
1800	return scnprintf(buf, PAGE_SIZE, "%d\n", msc->wrap);
1801}
1802
1803static ssize_t
1804wrap_store(struct device *dev, struct device_attribute *attr, const char *buf,
1805	   size_t size)
1806{
1807	struct msc *msc = dev_get_drvdata(dev);
1808	unsigned long val;
1809	int ret;
1810
1811	ret = kstrtoul(buf, 10, &val);
1812	if (ret)
1813		return ret;
1814
1815	msc->wrap = !!val;
1816
1817	return size;
1818}
1819
1820static DEVICE_ATTR_RW(wrap);
1821
1822static void msc_buffer_unassign(struct msc *msc)
1823{
1824	lockdep_assert_held(&msc->buf_mutex);
1825
1826	if (!msc->mbuf)
1827		return;
1828
1829	msc->mbuf->unassign(msc->mbuf_priv);
1830	msu_buffer_put(msc->mbuf);
1831	msc->mbuf_priv = NULL;
1832	msc->mbuf = NULL;
1833}
1834
1835static ssize_t
1836mode_show(struct device *dev, struct device_attribute *attr, char *buf)
1837{
1838	struct msc *msc = dev_get_drvdata(dev);
1839	const char *mode = msc_mode[msc->mode];
1840	ssize_t ret;
1841
1842	mutex_lock(&msc->buf_mutex);
1843	if (msc->mbuf)
1844		mode = msc->mbuf->name;
1845	ret = scnprintf(buf, PAGE_SIZE, "%s\n", mode);
1846	mutex_unlock(&msc->buf_mutex);
1847
1848	return ret;
1849}
1850
1851static ssize_t
1852mode_store(struct device *dev, struct device_attribute *attr, const char *buf,
1853	   size_t size)
1854{
1855	const struct msu_buffer *mbuf = NULL;
1856	struct msc *msc = dev_get_drvdata(dev);
1857	size_t len = size;
1858	char *cp, *mode;
1859	int i, ret;
1860
1861	if (!capable(CAP_SYS_RAWIO))
1862		return -EPERM;
1863
1864	cp = memchr(buf, '\n', len);
1865	if (cp)
1866		len = cp - buf;
1867
1868	mode = kstrndup(buf, len, GFP_KERNEL);
1869	if (!mode)
1870		return -ENOMEM;
1871
1872	i = match_string(msc_mode, ARRAY_SIZE(msc_mode), mode);
1873	if (i >= 0) {
1874		kfree(mode);
1875		goto found;
1876	}
1877
1878	/* Buffer sinks only work with a usable IRQ */
1879	if (!msc->do_irq) {
1880		kfree(mode);
1881		return -EINVAL;
1882	}
1883
1884	mbuf = msu_buffer_get(mode);
1885	kfree(mode);
1886	if (mbuf)
1887		goto found;
1888
1889	return -EINVAL;
1890
1891found:
1892	if (i == MSC_MODE_MULTI && msc->multi_is_broken)
1893		return -EOPNOTSUPP;
1894
1895	mutex_lock(&msc->buf_mutex);
1896	ret = 0;
1897
1898	/* Same buffer: do nothing */
1899	if (mbuf && mbuf == msc->mbuf) {
1900		/* put the extra reference we just got */
1901		msu_buffer_put(mbuf);
1902		goto unlock;
1903	}
1904
1905	ret = msc_buffer_unlocked_free_unless_used(msc);
1906	if (ret)
1907		goto unlock;
1908
1909	if (mbuf) {
1910		void *mbuf_priv = mbuf->assign(dev, &i);
1911
1912		if (!mbuf_priv) {
1913			ret = -ENOMEM;
1914			goto unlock;
1915		}
1916
1917		msc_buffer_unassign(msc);
1918		msc->mbuf_priv = mbuf_priv;
1919		msc->mbuf = mbuf;
1920	} else {
1921		msc_buffer_unassign(msc);
1922	}
1923
1924	msc->mode = i;
1925
1926unlock:
1927	if (ret && mbuf)
1928		msu_buffer_put(mbuf);
1929	mutex_unlock(&msc->buf_mutex);
1930
1931	return ret ? ret : size;
1932}
1933
1934static DEVICE_ATTR_RW(mode);
1935
1936static ssize_t
1937nr_pages_show(struct device *dev, struct device_attribute *attr, char *buf)
1938{
1939	struct msc *msc = dev_get_drvdata(dev);
1940	struct msc_window *win;
1941	size_t count = 0;
1942
1943	mutex_lock(&msc->buf_mutex);
1944
1945	if (msc->mode == MSC_MODE_SINGLE)
1946		count = scnprintf(buf, PAGE_SIZE, "%ld\n", msc->nr_pages);
1947	else if (msc->mode == MSC_MODE_MULTI) {
1948		list_for_each_entry(win, &msc->win_list, entry) {
1949			count += scnprintf(buf + count, PAGE_SIZE - count,
1950					   "%d%c", win->nr_blocks,
1951					   msc_is_last_win(win) ? '\n' : ',');
1952		}
1953	} else {
1954		count = scnprintf(buf, PAGE_SIZE, "unsupported\n");
1955	}
1956
1957	mutex_unlock(&msc->buf_mutex);
1958
1959	return count;
1960}
1961
1962static ssize_t
1963nr_pages_store(struct device *dev, struct device_attribute *attr,
1964	       const char *buf, size_t size)
1965{
1966	struct msc *msc = dev_get_drvdata(dev);
1967	unsigned long val, *win = NULL, *rewin;
1968	size_t len = size;
1969	const char *p = buf;
1970	char *end, *s;
1971	int ret, nr_wins = 0;
1972
1973	if (!capable(CAP_SYS_RAWIO))
1974		return -EPERM;
1975
1976	ret = msc_buffer_free_unless_used(msc);
1977	if (ret)
1978		return ret;
1979
1980	/* scan the comma-separated list of allocation sizes */
1981	end = memchr(buf, '\n', len);
1982	if (end)
1983		len = end - buf;
1984
1985	do {
1986		end = memchr(p, ',', len);
1987		s = kstrndup(p, end ? end - p : len, GFP_KERNEL);
1988		if (!s) {
1989			ret = -ENOMEM;
1990			goto free_win;
1991		}
1992
1993		ret = kstrtoul(s, 10, &val);
1994		kfree(s);
1995
1996		if (ret || !val)
1997			goto free_win;
1998
1999		if (nr_wins && msc->mode == MSC_MODE_SINGLE) {
2000			ret = -EINVAL;
2001			goto free_win;
2002		}
2003
2004		nr_wins++;
2005		rewin = krealloc(win, sizeof(*win) * nr_wins, GFP_KERNEL);
2006		if (!rewin) {
2007			kfree(win);
2008			return -ENOMEM;
2009		}
2010
2011		win = rewin;
2012		win[nr_wins - 1] = val;
2013
2014		if (!end)
2015			break;
2016
2017		/* consume the number and the following comma, hence +1 */
2018		len -= end - p + 1;
2019		p = end + 1;
2020	} while (len);
2021
2022	mutex_lock(&msc->buf_mutex);
2023	ret = msc_buffer_alloc(msc, win, nr_wins);
2024	mutex_unlock(&msc->buf_mutex);
2025
2026free_win:
2027	kfree(win);
2028
2029	return ret ? ret : size;
2030}
2031
2032static DEVICE_ATTR_RW(nr_pages);
2033
2034static ssize_t
2035win_switch_store(struct device *dev, struct device_attribute *attr,
2036		 const char *buf, size_t size)
2037{
2038	struct msc *msc = dev_get_drvdata(dev);
2039	unsigned long val;
2040	int ret;
2041
2042	ret = kstrtoul(buf, 10, &val);
2043	if (ret)
2044		return ret;
2045
2046	if (val != 1)
2047		return -EINVAL;
2048
2049	ret = -EINVAL;
2050	mutex_lock(&msc->buf_mutex);
2051	/*
2052	 * Window switch can only happen in the "multi" mode.
2053	 * If a external buffer is engaged, they have the full
2054	 * control over window switching.
2055	 */
2056	if (msc->mode == MSC_MODE_MULTI && !msc->mbuf)
2057		ret = msc_win_switch(msc);
 
 
2058	mutex_unlock(&msc->buf_mutex);
2059
2060	return ret ? ret : size;
2061}
2062
2063static DEVICE_ATTR_WO(win_switch);
2064
2065static ssize_t stop_on_full_show(struct device *dev,
2066				 struct device_attribute *attr, char *buf)
2067{
2068	struct msc *msc = dev_get_drvdata(dev);
2069
2070	return sprintf(buf, "%d\n", msc->stop_on_full);
2071}
2072
2073static ssize_t stop_on_full_store(struct device *dev,
2074				  struct device_attribute *attr,
2075				  const char *buf, size_t size)
2076{
2077	struct msc *msc = dev_get_drvdata(dev);
2078	int ret;
2079
2080	ret = kstrtobool(buf, &msc->stop_on_full);
2081	if (ret)
2082		return ret;
2083
2084	return size;
2085}
2086
2087static DEVICE_ATTR_RW(stop_on_full);
2088
2089static struct attribute *msc_output_attrs[] = {
2090	&dev_attr_wrap.attr,
2091	&dev_attr_mode.attr,
2092	&dev_attr_nr_pages.attr,
2093	&dev_attr_win_switch.attr,
2094	&dev_attr_stop_on_full.attr,
2095	NULL,
2096};
2097
2098static struct attribute_group msc_output_group = {
2099	.attrs	= msc_output_attrs,
2100};
2101
2102static int intel_th_msc_probe(struct intel_th_device *thdev)
2103{
2104	struct device *dev = &thdev->dev;
2105	struct resource *res;
2106	struct msc *msc;
2107	void __iomem *base;
2108	int err;
2109
2110	res = intel_th_device_get_resource(thdev, IORESOURCE_MEM, 0);
2111	if (!res)
2112		return -ENODEV;
2113
2114	base = devm_ioremap(dev, res->start, resource_size(res));
2115	if (!base)
2116		return -ENOMEM;
2117
2118	msc = devm_kzalloc(dev, sizeof(*msc), GFP_KERNEL);
2119	if (!msc)
2120		return -ENOMEM;
2121
2122	res = intel_th_device_get_resource(thdev, IORESOURCE_IRQ, 1);
2123	if (!res)
2124		msc->do_irq = 1;
2125
2126	if (INTEL_TH_CAP(to_intel_th(thdev), multi_is_broken))
2127		msc->multi_is_broken = 1;
2128
2129	msc->index = thdev->id;
2130
2131	msc->thdev = thdev;
2132	msc->reg_base = base + msc->index * 0x100;
2133	msc->msu_base = base;
2134
2135	INIT_WORK(&msc->work, msc_work);
2136	err = intel_th_msc_init(msc);
2137	if (err)
2138		return err;
2139
2140	dev_set_drvdata(dev, msc);
2141
2142	return 0;
2143}
2144
2145static void intel_th_msc_remove(struct intel_th_device *thdev)
2146{
2147	struct msc *msc = dev_get_drvdata(&thdev->dev);
2148	int ret;
2149
2150	intel_th_msc_deactivate(thdev);
2151
2152	/*
2153	 * Buffers should not be used at this point except if the
2154	 * output character device is still open and the parent
2155	 * device gets detached from its bus, which is a FIXME.
2156	 */
2157	ret = msc_buffer_free_unless_used(msc);
2158	WARN_ON_ONCE(ret);
2159}
2160
2161static struct intel_th_driver intel_th_msc_driver = {
2162	.probe	= intel_th_msc_probe,
2163	.remove	= intel_th_msc_remove,
2164	.irq		= intel_th_msc_interrupt,
2165	.wait_empty	= intel_th_msc_wait_empty,
2166	.activate	= intel_th_msc_activate,
2167	.deactivate	= intel_th_msc_deactivate,
2168	.fops	= &intel_th_msc_fops,
2169	.attr_group	= &msc_output_group,
2170	.driver	= {
2171		.name	= "msc",
2172		.owner	= THIS_MODULE,
2173	},
2174};
2175
2176module_driver(intel_th_msc_driver,
2177	      intel_th_driver_register,
2178	      intel_th_driver_unregister);
2179
2180MODULE_LICENSE("GPL v2");
2181MODULE_DESCRIPTION("Intel(R) Trace Hub Memory Storage Unit driver");
2182MODULE_AUTHOR("Alexander Shishkin <alexander.shishkin@linux.intel.com>");