Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0-only OR MIT
   2/* Copyright (c) 2023 Imagination Technologies Ltd. */
   3
   4#include "pvr_mmu.h"
   5
   6#include "pvr_ccb.h"
   7#include "pvr_device.h"
   8#include "pvr_fw.h"
   9#include "pvr_gem.h"
  10#include "pvr_power.h"
  11#include "pvr_rogue_fwif.h"
  12#include "pvr_rogue_mmu_defs.h"
  13
  14#include <drm/drm_drv.h>
  15#include <linux/atomic.h>
  16#include <linux/bitops.h>
  17#include <linux/dma-mapping.h>
  18#include <linux/kmemleak.h>
  19#include <linux/minmax.h>
  20#include <linux/sizes.h>
  21
  22#define PVR_SHIFT_FROM_SIZE(size_) (__builtin_ctzll(size_))
  23#define PVR_MASK_FROM_SIZE(size_) (~((size_) - U64_C(1)))
  24
  25/*
  26 * The value of the device page size (%PVR_DEVICE_PAGE_SIZE) is currently
  27 * pegged to the host page size (%PAGE_SIZE). This chunk of macro goodness both
  28 * ensures that the selected host page size corresponds to a valid device page
  29 * size and sets up values needed by the MMU code below.
  30 */
  31#if (PVR_DEVICE_PAGE_SIZE == SZ_4K)
  32# define ROGUE_MMUCTRL_PAGE_SIZE_X ROGUE_MMUCTRL_PAGE_SIZE_4KB
  33# define ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT ROGUE_MMUCTRL_PAGE_4KB_RANGE_SHIFT
  34# define ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK ROGUE_MMUCTRL_PAGE_4KB_RANGE_CLRMSK
  35#elif (PVR_DEVICE_PAGE_SIZE == SZ_16K)
  36# define ROGUE_MMUCTRL_PAGE_SIZE_X ROGUE_MMUCTRL_PAGE_SIZE_16KB
  37# define ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT ROGUE_MMUCTRL_PAGE_16KB_RANGE_SHIFT
  38# define ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK ROGUE_MMUCTRL_PAGE_16KB_RANGE_CLRMSK
  39#elif (PVR_DEVICE_PAGE_SIZE == SZ_64K)
  40# define ROGUE_MMUCTRL_PAGE_SIZE_X ROGUE_MMUCTRL_PAGE_SIZE_64KB
  41# define ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT ROGUE_MMUCTRL_PAGE_64KB_RANGE_SHIFT
  42# define ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK ROGUE_MMUCTRL_PAGE_64KB_RANGE_CLRMSK
  43#elif (PVR_DEVICE_PAGE_SIZE == SZ_256K)
  44# define ROGUE_MMUCTRL_PAGE_SIZE_X ROGUE_MMUCTRL_PAGE_SIZE_256KB
  45# define ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT ROGUE_MMUCTRL_PAGE_256KB_RANGE_SHIFT
  46# define ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK ROGUE_MMUCTRL_PAGE_256KB_RANGE_CLRMSK
  47#elif (PVR_DEVICE_PAGE_SIZE == SZ_1M)
  48# define ROGUE_MMUCTRL_PAGE_SIZE_X ROGUE_MMUCTRL_PAGE_SIZE_1MB
  49# define ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT ROGUE_MMUCTRL_PAGE_1MB_RANGE_SHIFT
  50# define ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK ROGUE_MMUCTRL_PAGE_1MB_RANGE_CLRMSK
  51#elif (PVR_DEVICE_PAGE_SIZE == SZ_2M)
  52# define ROGUE_MMUCTRL_PAGE_SIZE_X ROGUE_MMUCTRL_PAGE_SIZE_2MB
  53# define ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT ROGUE_MMUCTRL_PAGE_2MB_RANGE_SHIFT
  54# define ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK ROGUE_MMUCTRL_PAGE_2MB_RANGE_CLRMSK
  55#else
  56# error Unsupported device page size PVR_DEVICE_PAGE_SIZE
  57#endif
  58
  59#define ROGUE_MMUCTRL_ENTRIES_PT_VALUE_X   \
  60	(ROGUE_MMUCTRL_ENTRIES_PT_VALUE >> \
  61	 (PVR_DEVICE_PAGE_SHIFT - PVR_SHIFT_FROM_SIZE(SZ_4K)))
  62
  63enum pvr_mmu_sync_level {
  64	PVR_MMU_SYNC_LEVEL_NONE = -1,
  65	PVR_MMU_SYNC_LEVEL_0 = 0,
  66	PVR_MMU_SYNC_LEVEL_1 = 1,
  67	PVR_MMU_SYNC_LEVEL_2 = 2,
  68};
  69
  70#define PVR_MMU_SYNC_LEVEL_0_FLAGS (ROGUE_FWIF_MMUCACHEDATA_FLAGS_PT | \
  71				    ROGUE_FWIF_MMUCACHEDATA_FLAGS_INTERRUPT | \
  72				    ROGUE_FWIF_MMUCACHEDATA_FLAGS_TLB)
  73#define PVR_MMU_SYNC_LEVEL_1_FLAGS (PVR_MMU_SYNC_LEVEL_0_FLAGS | ROGUE_FWIF_MMUCACHEDATA_FLAGS_PD)
  74#define PVR_MMU_SYNC_LEVEL_2_FLAGS (PVR_MMU_SYNC_LEVEL_1_FLAGS | ROGUE_FWIF_MMUCACHEDATA_FLAGS_PC)
  75
  76/**
  77 * pvr_mmu_set_flush_flags() - Set MMU cache flush flags for next call to
  78 *                             pvr_mmu_flush_exec().
  79 * @pvr_dev: Target PowerVR device.
  80 * @flags: MMU flush flags. Must be one of %PVR_MMU_SYNC_LEVEL_*_FLAGS.
  81 *
  82 * This function must be called following any possible change to the MMU page
  83 * tables.
  84 */
  85static void pvr_mmu_set_flush_flags(struct pvr_device *pvr_dev, u32 flags)
  86{
  87	atomic_fetch_or(flags, &pvr_dev->mmu_flush_cache_flags);
  88}
  89
  90/**
  91 * pvr_mmu_flush_request_all() - Request flush of all MMU caches when
  92 * subsequently calling pvr_mmu_flush_exec().
  93 * @pvr_dev: Target PowerVR device.
  94 *
  95 * This function must be called following any possible change to the MMU page
  96 * tables.
  97 */
  98void pvr_mmu_flush_request_all(struct pvr_device *pvr_dev)
  99{
 100	pvr_mmu_set_flush_flags(pvr_dev, PVR_MMU_SYNC_LEVEL_2_FLAGS);
 101}
 102
 103/**
 104 * pvr_mmu_flush_exec() - Execute a flush of all MMU caches previously
 105 * requested.
 106 * @pvr_dev: Target PowerVR device.
 107 * @wait: Do not return until the flush is completed.
 108 *
 109 * This function must be called prior to submitting any new GPU job. The flush
 110 * will complete before the jobs are scheduled, so this can be called once after
 111 * a series of maps. However, a single unmap should always be immediately
 112 * followed by a flush and it should be explicitly waited by setting @wait.
 113 *
 114 * As a failure to flush the MMU caches could risk memory corruption, if the
 115 * flush fails (implying the firmware is not responding) then the GPU device is
 116 * marked as lost.
 117 *
 118 * Returns:
 119 *  * 0 on success when @wait is true, or
 120 *  * -%EIO if the device is unavailable, or
 121 *  * Any error encountered while submitting the flush command via the KCCB.
 122 */
 123int pvr_mmu_flush_exec(struct pvr_device *pvr_dev, bool wait)
 124{
 125	struct rogue_fwif_kccb_cmd cmd_mmu_cache = {};
 126	struct rogue_fwif_mmucachedata *cmd_mmu_cache_data =
 127		&cmd_mmu_cache.cmd_data.mmu_cache_data;
 128	int err = 0;
 129	u32 slot;
 130	int idx;
 131
 132	if (!drm_dev_enter(from_pvr_device(pvr_dev), &idx))
 133		return -EIO;
 134
 135	/* Can't flush MMU if the firmware hasn't booted yet. */
 136	if (!pvr_dev->fw_dev.booted)
 137		goto err_drm_dev_exit;
 138
 139	cmd_mmu_cache_data->cache_flags =
 140		atomic_xchg(&pvr_dev->mmu_flush_cache_flags, 0);
 141
 142	if (!cmd_mmu_cache_data->cache_flags)
 143		goto err_drm_dev_exit;
 144
 145	cmd_mmu_cache.cmd_type = ROGUE_FWIF_KCCB_CMD_MMUCACHE;
 146
 147	pvr_fw_object_get_fw_addr(pvr_dev->fw_dev.mem.mmucache_sync_obj,
 148				  &cmd_mmu_cache_data->mmu_cache_sync_fw_addr);
 149	cmd_mmu_cache_data->mmu_cache_sync_update_value = 0;
 150
 151	err = pvr_kccb_send_cmd(pvr_dev, &cmd_mmu_cache, &slot);
 152	if (err)
 153		goto err_reset_and_retry;
 154
 155	err = pvr_kccb_wait_for_completion(pvr_dev, slot, HZ, NULL);
 156	if (err)
 157		goto err_reset_and_retry;
 158
 159	drm_dev_exit(idx);
 160
 161	return 0;
 162
 163err_reset_and_retry:
 164	/*
 165	 * Flush command failure is most likely the result of a firmware lockup. Hard
 166	 * reset the GPU and retry.
 167	 */
 168	err = pvr_power_reset(pvr_dev, true);
 169	if (err)
 170		goto err_drm_dev_exit; /* Device is lost. */
 171
 172	/* Retry sending flush request. */
 173	err = pvr_kccb_send_cmd(pvr_dev, &cmd_mmu_cache, &slot);
 174	if (err) {
 175		pvr_device_lost(pvr_dev);
 176		goto err_drm_dev_exit;
 177	}
 178
 179	if (wait) {
 180		err = pvr_kccb_wait_for_completion(pvr_dev, slot, HZ, NULL);
 181		if (err)
 182			pvr_device_lost(pvr_dev);
 183	}
 184
 185err_drm_dev_exit:
 186	drm_dev_exit(idx);
 187
 188	return err;
 189}
 190
 191/**
 192 * DOC: PowerVR Virtual Memory Handling
 193 */
 194/**
 195 * DOC: PowerVR Virtual Memory Handling (constants)
 196 *
 197 * .. c:macro:: PVR_IDX_INVALID
 198 *
 199 *    Default value for a u16-based index.
 200 *
 201 *    This value cannot be zero, since zero is a valid index value.
 202 */
 203#define PVR_IDX_INVALID ((u16)(-1))
 204
 205/**
 206 * DOC: MMU backing pages
 207 */
 208/**
 209 * DOC: MMU backing pages (constants)
 210 *
 211 * .. c:macro:: PVR_MMU_BACKING_PAGE_SIZE
 212 *
 213 *    Page size of a PowerVR device's integrated MMU. The CPU page size must be
 214 *    at least as large as this value for the current implementation; this is
 215 *    checked at compile-time.
 216 */
 217#define PVR_MMU_BACKING_PAGE_SIZE SZ_4K
 218static_assert(PAGE_SIZE >= PVR_MMU_BACKING_PAGE_SIZE);
 219
 220/**
 221 * struct pvr_mmu_backing_page - Represents a single page used to back a page
 222 *                              table of any level.
 223 * @dma_addr: DMA address of this page.
 224 * @host_ptr: CPU address of this page.
 225 * @pvr_dev: The PowerVR device to which this page is associated. **For
 226 *           internal use only.**
 227 */
 228struct pvr_mmu_backing_page {
 229	dma_addr_t dma_addr;
 230	void *host_ptr;
 231/* private: internal use only */
 232	struct page *raw_page;
 233	struct pvr_device *pvr_dev;
 234};
 235
 236/**
 237 * pvr_mmu_backing_page_init() - Initialize a MMU backing page.
 238 * @page: Target backing page.
 239 * @pvr_dev: Target PowerVR device.
 240 *
 241 * This function performs three distinct operations:
 242 *
 243 * 1. Allocate a single page,
 244 * 2. Map the page to the CPU, and
 245 * 3. Map the page to DMA-space.
 246 *
 247 * It is expected that @page be zeroed (e.g. from kzalloc()) before calling
 248 * this function.
 249 *
 250 * Return:
 251 *  * 0 on success, or
 252 *  * -%ENOMEM if allocation of the backing page or mapping of the backing
 253 *    page to DMA fails.
 254 */
 255static int
 256pvr_mmu_backing_page_init(struct pvr_mmu_backing_page *page,
 257			  struct pvr_device *pvr_dev)
 258{
 259	struct device *dev = from_pvr_device(pvr_dev)->dev;
 260
 261	struct page *raw_page;
 262	int err;
 263
 264	dma_addr_t dma_addr;
 265	void *host_ptr;
 266
 267	raw_page = alloc_page(__GFP_ZERO | GFP_KERNEL);
 268	if (!raw_page)
 269		return -ENOMEM;
 270
 271	host_ptr = vmap(&raw_page, 1, VM_MAP, pgprot_writecombine(PAGE_KERNEL));
 272	if (!host_ptr) {
 273		err = -ENOMEM;
 274		goto err_free_page;
 275	}
 276
 277	dma_addr = dma_map_page(dev, raw_page, 0, PVR_MMU_BACKING_PAGE_SIZE,
 278				DMA_TO_DEVICE);
 279	if (dma_mapping_error(dev, dma_addr)) {
 280		err = -ENOMEM;
 281		goto err_unmap_page;
 282	}
 283
 284	page->dma_addr = dma_addr;
 285	page->host_ptr = host_ptr;
 286	page->pvr_dev = pvr_dev;
 287	page->raw_page = raw_page;
 288	kmemleak_alloc(page->host_ptr, PAGE_SIZE, 1, GFP_KERNEL);
 289
 290	return 0;
 291
 292err_unmap_page:
 293	vunmap(host_ptr);
 294
 295err_free_page:
 296	__free_page(raw_page);
 297
 298	return err;
 299}
 300
 301/**
 302 * pvr_mmu_backing_page_fini() - Teardown a MMU backing page.
 303 * @page: Target backing page.
 304 *
 305 * This function performs the mirror operations to pvr_mmu_backing_page_init(),
 306 * in reverse order:
 307 *
 308 * 1. Unmap the page from DMA-space,
 309 * 2. Unmap the page from the CPU, and
 310 * 3. Free the page.
 311 *
 312 * It also zeros @page.
 313 *
 314 * It is a no-op to call this function a second (or further) time on any @page.
 315 */
 316static void
 317pvr_mmu_backing_page_fini(struct pvr_mmu_backing_page *page)
 318{
 319	struct device *dev;
 320
 321	/* Do nothing if no allocation is present. */
 322	if (!page->pvr_dev)
 323		return;
 324
 325	dev = from_pvr_device(page->pvr_dev)->dev;
 326
 327	dma_unmap_page(dev, page->dma_addr, PVR_MMU_BACKING_PAGE_SIZE,
 328		       DMA_TO_DEVICE);
 329
 330	kmemleak_free(page->host_ptr);
 331	vunmap(page->host_ptr);
 332
 333	__free_page(page->raw_page);
 334
 335	memset(page, 0, sizeof(*page));
 336}
 337
 338/**
 339 * pvr_mmu_backing_page_sync() - Flush a MMU backing page from the CPU to the
 340 * device.
 341 * @page: Target backing page.
 342 * @flags: MMU flush flags. Must be one of %PVR_MMU_SYNC_LEVEL_*_FLAGS.
 343 *
 344 * .. caution::
 345 *
 346 *    **This is potentially an expensive function call.** Only call
 347 *    pvr_mmu_backing_page_sync() once you're sure you have no more changes to
 348 *    make to the backing page in the immediate future.
 349 */
 350static void
 351pvr_mmu_backing_page_sync(struct pvr_mmu_backing_page *page, u32 flags)
 352{
 353	struct pvr_device *pvr_dev = page->pvr_dev;
 354	struct device *dev;
 355
 356	/*
 357	 * Do nothing if no allocation is present. This may be the case if
 358	 * we are unmapping pages.
 359	 */
 360	if (!pvr_dev)
 361		return;
 362
 363	dev = from_pvr_device(pvr_dev)->dev;
 364
 365	dma_sync_single_for_device(dev, page->dma_addr,
 366				   PVR_MMU_BACKING_PAGE_SIZE, DMA_TO_DEVICE);
 367
 368	pvr_mmu_set_flush_flags(pvr_dev, flags);
 369}
 370
 371/**
 372 * DOC: Raw page tables
 373 */
 374
 375#define PVR_PAGE_TABLE_TYPEOF_ENTRY(level_) \
 376	typeof_member(struct pvr_page_table_l##level_##_entry_raw, val)
 377
 378#define PVR_PAGE_TABLE_FIELD_GET(level_, name_, field_, entry_)           \
 379	(((entry_).val &                                           \
 380	  ~ROGUE_MMUCTRL_##name_##_DATA_##field_##_CLRMSK) >> \
 381	 ROGUE_MMUCTRL_##name_##_DATA_##field_##_SHIFT)
 382
 383#define PVR_PAGE_TABLE_FIELD_PREP(level_, name_, field_, val_)            \
 384	((((PVR_PAGE_TABLE_TYPEOF_ENTRY(level_))(val_))            \
 385	  << ROGUE_MMUCTRL_##name_##_DATA_##field_##_SHIFT) & \
 386	 ~ROGUE_MMUCTRL_##name_##_DATA_##field_##_CLRMSK)
 387
 388/**
 389 * struct pvr_page_table_l2_entry_raw - A single entry in a level 2 page table.
 390 * @val: The raw value of this entry.
 391 *
 392 * This type is a structure for type-checking purposes. At compile-time, its
 393 * size is checked against %ROGUE_MMUCTRL_ENTRY_SIZE_PC_VALUE.
 394 *
 395 * The value stored in this structure can be decoded using the following bitmap:
 396 *
 397 * .. flat-table::
 398 *    :widths: 1 5
 399 *    :stub-columns: 1
 400 *
 401 *    * - 31..4
 402 *      - **Level 1 Page Table Base Address:** Bits 39..12 of the L1
 403 *        page table base address, which is 4KiB aligned.
 404 *
 405 *    * - 3..2
 406 *      - *(reserved)*
 407 *
 408 *    * - 1
 409 *      - **Pending:** When valid bit is not set, indicates that a valid
 410 *        entry is pending and the MMU should wait for the driver to map
 411 *        the entry. This is used to support page demand mapping of
 412 *        memory.
 413 *
 414 *    * - 0
 415 *      - **Valid:** Indicates that the entry contains a valid L1 page
 416 *        table. If the valid bit is not set, then an attempted use of
 417 *        the page would result in a page fault.
 418 */
 419struct pvr_page_table_l2_entry_raw {
 420	u32 val;
 421} __packed;
 422static_assert(sizeof(struct pvr_page_table_l2_entry_raw) * 8 ==
 423	      ROGUE_MMUCTRL_ENTRY_SIZE_PC_VALUE);
 424
 425static bool
 426pvr_page_table_l2_entry_raw_is_valid(struct pvr_page_table_l2_entry_raw entry)
 427{
 428	return PVR_PAGE_TABLE_FIELD_GET(2, PC, VALID, entry);
 429}
 430
 431/**
 432 * pvr_page_table_l2_entry_raw_set() - Write a valid entry into a raw level 2
 433 *                                     page table.
 434 * @entry: Target raw level 2 page table entry.
 435 * @child_table_dma_addr: DMA address of the level 1 page table to be
 436 *                        associated with @entry.
 437 *
 438 * When calling this function, @child_table_dma_addr must be a valid DMA
 439 * address and a multiple of %ROGUE_MMUCTRL_PC_DATA_PD_BASE_ALIGNSIZE.
 440 */
 441static void
 442pvr_page_table_l2_entry_raw_set(struct pvr_page_table_l2_entry_raw *entry,
 443				dma_addr_t child_table_dma_addr)
 444{
 445	child_table_dma_addr >>= ROGUE_MMUCTRL_PC_DATA_PD_BASE_ALIGNSHIFT;
 446
 447	WRITE_ONCE(entry->val,
 448		   PVR_PAGE_TABLE_FIELD_PREP(2, PC, VALID, true) |
 449		   PVR_PAGE_TABLE_FIELD_PREP(2, PC, ENTRY_PENDING, false) |
 450		   PVR_PAGE_TABLE_FIELD_PREP(2, PC, PD_BASE, child_table_dma_addr));
 451}
 452
 453static void
 454pvr_page_table_l2_entry_raw_clear(struct pvr_page_table_l2_entry_raw *entry)
 455{
 456	WRITE_ONCE(entry->val, 0);
 457}
 458
 459/**
 460 * struct pvr_page_table_l1_entry_raw - A single entry in a level 1 page table.
 461 * @val: The raw value of this entry.
 462 *
 463 * This type is a structure for type-checking purposes. At compile-time, its
 464 * size is checked against %ROGUE_MMUCTRL_ENTRY_SIZE_PD_VALUE.
 465 *
 466 * The value stored in this structure can be decoded using the following bitmap:
 467 *
 468 * .. flat-table::
 469 *    :widths: 1 5
 470 *    :stub-columns: 1
 471 *
 472 *    * - 63..41
 473 *      - *(reserved)*
 474 *
 475 *    * - 40
 476 *      - **Pending:** When valid bit is not set, indicates that a valid entry
 477 *        is pending and the MMU should wait for the driver to map the entry.
 478 *        This is used to support page demand mapping of memory.
 479 *
 480 *    * - 39..5
 481 *      - **Level 0 Page Table Base Address:** The way this value is
 482 *        interpreted depends on the page size. Bits not specified in the
 483 *        table below (e.g. bits 11..5 for page size 4KiB) should be
 484 *        considered reserved.
 485 *
 486 *        This table shows the bits used in an L1 page table entry to
 487 *        represent the Physical Table Base Address for a given Page Size.
 488 *        Since each L1 page table entry covers 2MiB of address space, the
 489 *        maximum page size is 2MiB.
 490 *
 491 *        .. flat-table::
 492 *           :widths: 1 1 1 1
 493 *           :header-rows: 1
 494 *           :stub-columns: 1
 495 *
 496 *           * - Page size
 497 *             - L0 page table base address bits
 498 *             - Number of L0 page table entries
 499 *             - Size of L0 page table
 500 *
 501 *           * - 4KiB
 502 *             - 39..12
 503 *             - 512
 504 *             - 4KiB
 505 *
 506 *           * - 16KiB
 507 *             - 39..10
 508 *             - 128
 509 *             - 1KiB
 510 *
 511 *           * - 64KiB
 512 *             - 39..8
 513 *             - 32
 514 *             - 256B
 515 *
 516 *           * - 256KiB
 517 *             - 39..6
 518 *             - 8
 519 *             - 64B
 520 *
 521 *           * - 1MiB
 522 *             - 39..5 (4 = '0')
 523 *             - 2
 524 *             - 16B
 525 *
 526 *           * - 2MiB
 527 *             - 39..5 (4..3 = '00')
 528 *             - 1
 529 *             - 8B
 530 *
 531 *    * - 4
 532 *      - *(reserved)*
 533 *
 534 *    * - 3..1
 535 *      - **Page Size:** Sets the page size, from 4KiB to 2MiB.
 536 *
 537 *    * - 0
 538 *      - **Valid:** Indicates that the entry contains a valid L0 page table.
 539 *        If the valid bit is not set, then an attempted use of the page would
 540 *        result in a page fault.
 541 */
 542struct pvr_page_table_l1_entry_raw {
 543	u64 val;
 544} __packed;
 545static_assert(sizeof(struct pvr_page_table_l1_entry_raw) * 8 ==
 546	      ROGUE_MMUCTRL_ENTRY_SIZE_PD_VALUE);
 547
 548static bool
 549pvr_page_table_l1_entry_raw_is_valid(struct pvr_page_table_l1_entry_raw entry)
 550{
 551	return PVR_PAGE_TABLE_FIELD_GET(1, PD, VALID, entry);
 552}
 553
 554/**
 555 * pvr_page_table_l1_entry_raw_set() - Write a valid entry into a raw level 1
 556 *                                     page table.
 557 * @entry: Target raw level 1 page table entry.
 558 * @child_table_dma_addr: DMA address of the level 0 page table to be
 559 *                        associated with @entry.
 560 *
 561 * When calling this function, @child_table_dma_addr must be a valid DMA
 562 * address and a multiple of 4 KiB.
 563 */
 564static void
 565pvr_page_table_l1_entry_raw_set(struct pvr_page_table_l1_entry_raw *entry,
 566				dma_addr_t child_table_dma_addr)
 567{
 568	WRITE_ONCE(entry->val,
 569		   PVR_PAGE_TABLE_FIELD_PREP(1, PD, VALID, true) |
 570		   PVR_PAGE_TABLE_FIELD_PREP(1, PD, ENTRY_PENDING, false) |
 571		   PVR_PAGE_TABLE_FIELD_PREP(1, PD, PAGE_SIZE, ROGUE_MMUCTRL_PAGE_SIZE_X) |
 572		   /*
 573		    * The use of a 4K-specific macro here is correct. It is
 574		    * a future optimization to allocate sub-host-page-sized
 575		    * blocks for individual tables, so the condition that any
 576		    * page table address is aligned to the size of the
 577		    * largest (a 4KB) table currently holds.
 578		    */
 579		   (child_table_dma_addr & ~ROGUE_MMUCTRL_PT_BASE_4KB_RANGE_CLRMSK));
 580}
 581
 582static void
 583pvr_page_table_l1_entry_raw_clear(struct pvr_page_table_l1_entry_raw *entry)
 584{
 585	WRITE_ONCE(entry->val, 0);
 586}
 587
 588/**
 589 * struct pvr_page_table_l0_entry_raw - A single entry in a level 0 page table.
 590 * @val: The raw value of this entry.
 591 *
 592 * This type is a structure for type-checking purposes. At compile-time, its
 593 * size is checked against %ROGUE_MMUCTRL_ENTRY_SIZE_PT_VALUE.
 594 *
 595 * The value stored in this structure can be decoded using the following bitmap:
 596 *
 597 * .. flat-table::
 598 *    :widths: 1 5
 599 *    :stub-columns: 1
 600 *
 601 *    * - 63
 602 *      - *(reserved)*
 603 *
 604 *    * - 62
 605 *      - **PM/FW Protect:** Indicates a protected region which only the
 606 *        Parameter Manager (PM) or firmware processor can write to.
 607 *
 608 *    * - 61..40
 609 *      - **VP Page (High):** Virtual-physical page used for Parameter Manager
 610 *        (PM) memory. This field is only used if the additional level of PB
 611 *        virtualization is enabled. The VP Page field is needed by the PM in
 612 *        order to correctly reconstitute the free lists after render
 613 *        completion. This (High) field holds bits 39..18 of the value; the
 614 *        Low field holds bits 17..12. Bits 11..0 are always zero because the
 615 *        value is always aligned to the 4KiB page size.
 616 *
 617 *    * - 39..12
 618 *      - **Physical Page Address:** The way this value is interpreted depends
 619 *        on the page size. Bits not specified in the table below (e.g. bits
 620 *        20..12 for page size 2MiB) should be considered reserved.
 621 *
 622 *        This table shows the bits used in an L0 page table entry to represent
 623 *        the Physical Page Address for a given page size (as defined in the
 624 *        associated L1 page table entry).
 625 *
 626 *        .. flat-table::
 627 *           :widths: 1 1
 628 *           :header-rows: 1
 629 *           :stub-columns: 1
 630 *
 631 *           * - Page size
 632 *             - Physical address bits
 633 *
 634 *           * - 4KiB
 635 *             - 39..12
 636 *
 637 *           * - 16KiB
 638 *             - 39..14
 639 *
 640 *           * - 64KiB
 641 *             - 39..16
 642 *
 643 *           * - 256KiB
 644 *             - 39..18
 645 *
 646 *           * - 1MiB
 647 *             - 39..20
 648 *
 649 *           * - 2MiB
 650 *             - 39..21
 651 *
 652 *    * - 11..6
 653 *      - **VP Page (Low):** Continuation of VP Page (High).
 654 *
 655 *    * - 5
 656 *      - **Pending:** When valid bit is not set, indicates that a valid entry
 657 *        is pending and the MMU should wait for the driver to map the entry.
 658 *        This is used to support page demand mapping of memory.
 659 *
 660 *    * - 4
 661 *      - **PM Src:** Set on Parameter Manager (PM) allocated page table
 662 *        entries when indicated by the PM. Note that this bit will only be set
 663 *        by the PM, not by the device driver.
 664 *
 665 *    * - 3
 666 *      - **SLC Bypass Control:** Specifies requests to this page should bypass
 667 *        the System Level Cache (SLC), if enabled in SLC configuration.
 668 *
 669 *    * - 2
 670 *      - **Cache Coherency:** Indicates that the page is coherent (i.e. it
 671 *        does not require a cache flush between operations on the CPU and the
 672 *        device).
 673 *
 674 *    * - 1
 675 *      - **Read Only:** If set, this bit indicates that the page is read only.
 676 *        An attempted write to this page would result in a write-protection
 677 *        fault.
 678 *
 679 *    * - 0
 680 *      - **Valid:** Indicates that the entry contains a valid page. If the
 681 *        valid bit is not set, then an attempted use of the page would result
 682 *        in a page fault.
 683 */
 684struct pvr_page_table_l0_entry_raw {
 685	u64 val;
 686} __packed;
 687static_assert(sizeof(struct pvr_page_table_l0_entry_raw) * 8 ==
 688	      ROGUE_MMUCTRL_ENTRY_SIZE_PT_VALUE);
 689
 690/**
 691 * struct pvr_page_flags_raw - The configurable flags from a single entry in a
 692 *                             level 0 page table.
 693 * @val: The raw value of these flags. Since these are a strict subset of
 694 *       &struct pvr_page_table_l0_entry_raw; use that type for our member here.
 695 *
 696 * The flags stored in this type are: PM/FW Protect; SLC Bypass Control; Cache
 697 * Coherency, and Read Only (bits 62, 3, 2 and 1 respectively).
 698 *
 699 * This type should never be instantiated directly; instead use
 700 * pvr_page_flags_raw_create() to ensure only valid bits of @val are set.
 701 */
 702struct pvr_page_flags_raw {
 703	struct pvr_page_table_l0_entry_raw val;
 704} __packed;
 705static_assert(sizeof(struct pvr_page_flags_raw) ==
 706	      sizeof(struct pvr_page_table_l0_entry_raw));
 707
 708static bool
 709pvr_page_table_l0_entry_raw_is_valid(struct pvr_page_table_l0_entry_raw entry)
 710{
 711	return PVR_PAGE_TABLE_FIELD_GET(0, PT, VALID, entry);
 712}
 713
 714/**
 715 * pvr_page_table_l0_entry_raw_set() - Write a valid entry into a raw level 0
 716 *                                     page table.
 717 * @entry: Target raw level 0 page table entry.
 718 * @dma_addr: DMA address of the physical page to be associated with @entry.
 719 * @flags: Options to be set on @entry.
 720 *
 721 * When calling this function, @child_table_dma_addr must be a valid DMA
 722 * address and a multiple of %PVR_DEVICE_PAGE_SIZE.
 723 *
 724 * The @flags parameter is directly assigned into @entry. It is the callers
 725 * responsibility to ensure that only bits specified in
 726 * &struct pvr_page_flags_raw are set in @flags.
 727 */
 728static void
 729pvr_page_table_l0_entry_raw_set(struct pvr_page_table_l0_entry_raw *entry,
 730				dma_addr_t dma_addr,
 731				struct pvr_page_flags_raw flags)
 732{
 733	WRITE_ONCE(entry->val, PVR_PAGE_TABLE_FIELD_PREP(0, PT, VALID, true) |
 734			       PVR_PAGE_TABLE_FIELD_PREP(0, PT, ENTRY_PENDING, false) |
 735			       (dma_addr & ~ROGUE_MMUCTRL_PAGE_X_RANGE_CLRMSK) |
 736			       flags.val.val);
 737}
 738
 739static void
 740pvr_page_table_l0_entry_raw_clear(struct pvr_page_table_l0_entry_raw *entry)
 741{
 742	WRITE_ONCE(entry->val, 0);
 743}
 744
 745/**
 746 * pvr_page_flags_raw_create() - Initialize the flag bits of a raw level 0 page
 747 *                               table entry.
 748 * @read_only: This page is read-only (see: Read Only).
 749 * @cache_coherent: This page does not require cache flushes (see: Cache
 750 *                  Coherency).
 751 * @slc_bypass: This page bypasses the device cache (see: SLC Bypass Control).
 752 * @pm_fw_protect: This page is only for use by the firmware or Parameter
 753 *                 Manager (see PM/FW Protect).
 754 *
 755 * For more details on the use of these four options, see their respective
 756 * entries in the table under &struct pvr_page_table_l0_entry_raw.
 757 *
 758 * Return:
 759 * A new &struct pvr_page_flags_raw instance which can be passed directly to
 760 * pvr_page_table_l0_entry_raw_set() or pvr_page_table_l0_insert().
 761 */
 762static struct pvr_page_flags_raw
 763pvr_page_flags_raw_create(bool read_only, bool cache_coherent, bool slc_bypass,
 764			  bool pm_fw_protect)
 765{
 766	struct pvr_page_flags_raw flags;
 767
 768	flags.val.val =
 769		PVR_PAGE_TABLE_FIELD_PREP(0, PT, READ_ONLY, read_only) |
 770		PVR_PAGE_TABLE_FIELD_PREP(0, PT, CC, cache_coherent) |
 771		PVR_PAGE_TABLE_FIELD_PREP(0, PT, SLC_BYPASS_CTRL, slc_bypass) |
 772		PVR_PAGE_TABLE_FIELD_PREP(0, PT, PM_META_PROTECT, pm_fw_protect);
 773
 774	return flags;
 775}
 776
 777/**
 778 * struct pvr_page_table_l2_raw - The raw data of a level 2 page table.
 779 *
 780 * This type is a structure for type-checking purposes. At compile-time, its
 781 * size is checked against %PVR_MMU_BACKING_PAGE_SIZE.
 782 */
 783struct pvr_page_table_l2_raw {
 784	/** @entries: The raw values of this table. */
 785	struct pvr_page_table_l2_entry_raw
 786		entries[ROGUE_MMUCTRL_ENTRIES_PC_VALUE];
 787} __packed;
 788static_assert(sizeof(struct pvr_page_table_l2_raw) == PVR_MMU_BACKING_PAGE_SIZE);
 789
 790/**
 791 * struct pvr_page_table_l1_raw - The raw data of a level 1 page table.
 792 *
 793 * This type is a structure for type-checking purposes. At compile-time, its
 794 * size is checked against %PVR_MMU_BACKING_PAGE_SIZE.
 795 */
 796struct pvr_page_table_l1_raw {
 797	/** @entries: The raw values of this table. */
 798	struct pvr_page_table_l1_entry_raw
 799		entries[ROGUE_MMUCTRL_ENTRIES_PD_VALUE];
 800} __packed;
 801static_assert(sizeof(struct pvr_page_table_l1_raw) == PVR_MMU_BACKING_PAGE_SIZE);
 802
 803/**
 804 * struct pvr_page_table_l0_raw - The raw data of a level 0 page table.
 805 *
 806 * This type is a structure for type-checking purposes. At compile-time, its
 807 * size is checked against %PVR_MMU_BACKING_PAGE_SIZE.
 808 *
 809 * .. caution::
 810 *
 811 *    The size of level 0 page tables is variable depending on the page size
 812 *    specified in the associated level 1 page table entry. Since the device
 813 *    page size in use is pegged to the host page size, it cannot vary at
 814 *    runtime. This structure is therefore only defined to contain the required
 815 *    number of entries for the current device page size. **You should never
 816 *    read or write beyond the last supported entry.**
 817 */
 818struct pvr_page_table_l0_raw {
 819	/** @entries: The raw values of this table. */
 820	struct pvr_page_table_l0_entry_raw
 821		entries[ROGUE_MMUCTRL_ENTRIES_PT_VALUE_X];
 822} __packed;
 823static_assert(sizeof(struct pvr_page_table_l0_raw) <= PVR_MMU_BACKING_PAGE_SIZE);
 824
 825/**
 826 * DOC: Mirror page tables
 827 */
 828
 829/*
 830 * We pre-declare these types because they cross-depend on pointers to each
 831 * other.
 832 */
 833struct pvr_page_table_l1;
 834struct pvr_page_table_l0;
 835
 836/**
 837 * struct pvr_page_table_l2 - A wrapped level 2 page table.
 838 *
 839 * To access the raw part of this table, use pvr_page_table_l2_get_raw().
 840 * Alternatively to access a raw entry directly, use
 841 * pvr_page_table_l2_get_entry_raw().
 842 *
 843 * A level 2 page table forms the root of the page table tree structure, so
 844 * this type has no &parent or &parent_idx members.
 845 */
 846struct pvr_page_table_l2 {
 847	/**
 848	 * @entries: The children of this node in the page table tree
 849	 * structure. These are also mirror tables. The indexing of this array
 850	 * is identical to that of the raw equivalent
 851	 * (&pvr_page_table_l1_raw.entries).
 852	 */
 853	struct pvr_page_table_l1 *entries[ROGUE_MMUCTRL_ENTRIES_PC_VALUE];
 854
 855	/**
 856	 * @backing_page: A handle to the memory which holds the raw
 857	 * equivalent of this table. **For internal use only.**
 858	 */
 859	struct pvr_mmu_backing_page backing_page;
 860
 861	/**
 862	 * @entry_count: The current number of valid entries (that we know of)
 863	 * in this table. This value is essentially a refcount - the table is
 864	 * destroyed when this value is decremented to zero by
 865	 * pvr_page_table_l2_remove().
 866	 */
 867	u16 entry_count;
 868};
 869
 870/**
 871 * pvr_page_table_l2_init() - Initialize a level 2 page table.
 872 * @table: Target level 2 page table.
 873 * @pvr_dev: Target PowerVR device
 874 *
 875 * It is expected that @table be zeroed (e.g. from kzalloc()) before calling
 876 * this function.
 877 *
 878 * Return:
 879 *  * 0 on success, or
 880 *  * Any error encountered while intializing &table->backing_page using
 881 *    pvr_mmu_backing_page_init().
 882 */
 883static int
 884pvr_page_table_l2_init(struct pvr_page_table_l2 *table,
 885		       struct pvr_device *pvr_dev)
 886{
 887	return pvr_mmu_backing_page_init(&table->backing_page, pvr_dev);
 888}
 889
 890/**
 891 * pvr_page_table_l2_fini() - Teardown a level 2 page table.
 892 * @table: Target level 2 page table.
 893 *
 894 * It is an error to attempt to use @table after calling this function.
 895 */
 896static void
 897pvr_page_table_l2_fini(struct pvr_page_table_l2 *table)
 898{
 899	pvr_mmu_backing_page_fini(&table->backing_page);
 900}
 901
 902/**
 903 * pvr_page_table_l2_sync() - Flush a level 2 page table from the CPU to the
 904 *                            device.
 905 * @table: Target level 2 page table.
 906 *
 907 * This is just a thin wrapper around pvr_mmu_backing_page_sync(), so the
 908 * warning there applies here too: **Only call pvr_page_table_l2_sync() once
 909 * you're sure you have no more changes to make to** @table **in the immediate
 910 * future.**
 911 *
 912 * If child level 1 page tables of @table also need to be flushed, this should
 913 * be done first using pvr_page_table_l1_sync() *before* calling this function.
 914 */
 915static void
 916pvr_page_table_l2_sync(struct pvr_page_table_l2 *table)
 917{
 918	pvr_mmu_backing_page_sync(&table->backing_page, PVR_MMU_SYNC_LEVEL_2_FLAGS);
 919}
 920
 921/**
 922 * pvr_page_table_l2_get_raw() - Access the raw equivalent of a mirror level 2
 923 *                               page table.
 924 * @table: Target level 2 page table.
 925 *
 926 * Essentially returns the CPU address of the raw equivalent of @table, cast to
 927 * a &struct pvr_page_table_l2_raw pointer.
 928 *
 929 * You probably want to call pvr_page_table_l2_get_entry_raw() instead.
 930 *
 931 * Return:
 932 * The raw equivalent of @table.
 933 */
 934static struct pvr_page_table_l2_raw *
 935pvr_page_table_l2_get_raw(struct pvr_page_table_l2 *table)
 936{
 937	return table->backing_page.host_ptr;
 938}
 939
 940/**
 941 * pvr_page_table_l2_get_entry_raw() - Access an entry from the raw equivalent
 942 *                                     of a mirror level 2 page table.
 943 * @table: Target level 2 page table.
 944 * @idx: Index of the entry to access.
 945 *
 946 * Technically this function returns a pointer to a slot in a raw level 2 page
 947 * table, since the returned "entry" is not guaranteed to be valid. The caller
 948 * must verify the validity of the entry at the returned address (perhaps using
 949 * pvr_page_table_l2_entry_raw_is_valid()) before reading or overwriting it.
 950 *
 951 * The value of @idx is not checked here; it is the callers responsibility to
 952 * ensure @idx refers to a valid index within @table before dereferencing the
 953 * returned pointer.
 954 *
 955 * Return:
 956 * A pointer to the requested raw level 2 page table entry.
 957 */
 958static struct pvr_page_table_l2_entry_raw *
 959pvr_page_table_l2_get_entry_raw(struct pvr_page_table_l2 *table, u16 idx)
 960{
 961	return &pvr_page_table_l2_get_raw(table)->entries[idx];
 962}
 963
 964/**
 965 * pvr_page_table_l2_entry_is_valid() - Check if a level 2 page table entry is
 966 *                                      marked as valid.
 967 * @table: Target level 2 page table.
 968 * @idx: Index of the entry to check.
 969 *
 970 * The value of @idx is not checked here; it is the callers responsibility to
 971 * ensure @idx refers to a valid index within @table before calling this
 972 * function.
 973 */
 974static bool
 975pvr_page_table_l2_entry_is_valid(struct pvr_page_table_l2 *table, u16 idx)
 976{
 977	struct pvr_page_table_l2_entry_raw entry_raw =
 978		*pvr_page_table_l2_get_entry_raw(table, idx);
 979
 980	return pvr_page_table_l2_entry_raw_is_valid(entry_raw);
 981}
 982
 983/**
 984 * struct pvr_page_table_l1 - A wrapped level 1 page table.
 985 *
 986 * To access the raw part of this table, use pvr_page_table_l1_get_raw().
 987 * Alternatively to access a raw entry directly, use
 988 * pvr_page_table_l1_get_entry_raw().
 989 */
 990struct pvr_page_table_l1 {
 991	/**
 992	 * @entries: The children of this node in the page table tree
 993	 * structure. These are also mirror tables. The indexing of this array
 994	 * is identical to that of the raw equivalent
 995	 * (&pvr_page_table_l0_raw.entries).
 996	 */
 997	struct pvr_page_table_l0 *entries[ROGUE_MMUCTRL_ENTRIES_PD_VALUE];
 998
 999	/**
1000	 * @backing_page: A handle to the memory which holds the raw
1001	 * equivalent of this table. **For internal use only.**
1002	 */
1003	struct pvr_mmu_backing_page backing_page;
1004
1005	union {
1006		/**
1007		 * @parent: The parent of this node in the page table tree structure.
1008		 *
1009		 * This is also a mirror table.
1010		 *
1011		 * Only valid when the L1 page table is active. When the L1 page table
1012		 * has been removed and queued for destruction, the next_free field
1013		 * should be used instead.
1014		 */
1015		struct pvr_page_table_l2 *parent;
1016
1017		/**
1018		 * @next_free: Pointer to the next L1 page table to take/free.
1019		 *
1020		 * Used to form a linked list of L1 page tables. This is used
1021		 * when preallocating tables and when the page table has been
1022		 * removed and queued for destruction.
1023		 */
1024		struct pvr_page_table_l1 *next_free;
1025	};
1026
1027	/**
1028	 * @parent_idx: The index of the entry in the parent table (see
1029	 * @parent) which corresponds to this table.
1030	 */
1031	u16 parent_idx;
1032
1033	/**
1034	 * @entry_count: The current number of valid entries (that we know of)
1035	 * in this table. This value is essentially a refcount - the table is
1036	 * destroyed when this value is decremented to zero by
1037	 * pvr_page_table_l1_remove().
1038	 */
1039	u16 entry_count;
1040};
1041
1042/**
1043 * pvr_page_table_l1_init() - Initialize a level 1 page table.
1044 * @table: Target level 1 page table.
1045 * @pvr_dev: Target PowerVR device
1046 *
1047 * When this function returns successfully, @table is still not considered
1048 * valid. It must be inserted into the page table tree structure with
1049 * pvr_page_table_l2_insert() before it is ready for use.
1050 *
1051 * It is expected that @table be zeroed (e.g. from kzalloc()) before calling
1052 * this function.
1053 *
1054 * Return:
1055 *  * 0 on success, or
1056 *  * Any error encountered while intializing &table->backing_page using
1057 *    pvr_mmu_backing_page_init().
1058 */
1059static int
1060pvr_page_table_l1_init(struct pvr_page_table_l1 *table,
1061		       struct pvr_device *pvr_dev)
1062{
1063	table->parent_idx = PVR_IDX_INVALID;
1064
1065	return pvr_mmu_backing_page_init(&table->backing_page, pvr_dev);
1066}
1067
1068/**
1069 * pvr_page_table_l1_free() - Teardown a level 1 page table.
1070 * @table: Target level 1 page table.
1071 *
1072 * It is an error to attempt to use @table after calling this function, even
1073 * indirectly. This includes calling pvr_page_table_l2_remove(), which must
1074 * be called *before* pvr_page_table_l1_free().
1075 */
1076static void
1077pvr_page_table_l1_free(struct pvr_page_table_l1 *table)
1078{
1079	pvr_mmu_backing_page_fini(&table->backing_page);
1080	kfree(table);
1081}
1082
1083/**
1084 * pvr_page_table_l1_sync() - Flush a level 1 page table from the CPU to the
1085 *                            device.
1086 * @table: Target level 1 page table.
1087 *
1088 * This is just a thin wrapper around pvr_mmu_backing_page_sync(), so the
1089 * warning there applies here too: **Only call pvr_page_table_l1_sync() once
1090 * you're sure you have no more changes to make to** @table **in the immediate
1091 * future.**
1092 *
1093 * If child level 0 page tables of @table also need to be flushed, this should
1094 * be done first using pvr_page_table_l0_sync() *before* calling this function.
1095 */
1096static void
1097pvr_page_table_l1_sync(struct pvr_page_table_l1 *table)
1098{
1099	pvr_mmu_backing_page_sync(&table->backing_page, PVR_MMU_SYNC_LEVEL_1_FLAGS);
1100}
1101
1102/**
1103 * pvr_page_table_l1_get_raw() - Access the raw equivalent of a mirror level 1
1104 *                               page table.
1105 * @table: Target level 1 page table.
1106 *
1107 * Essentially returns the CPU address of the raw equivalent of @table, cast to
1108 * a &struct pvr_page_table_l1_raw pointer.
1109 *
1110 * You probably want to call pvr_page_table_l1_get_entry_raw() instead.
1111 *
1112 * Return:
1113 * The raw equivalent of @table.
1114 */
1115static struct pvr_page_table_l1_raw *
1116pvr_page_table_l1_get_raw(struct pvr_page_table_l1 *table)
1117{
1118	return table->backing_page.host_ptr;
1119}
1120
1121/**
1122 * pvr_page_table_l1_get_entry_raw() - Access an entry from the raw equivalent
1123 *                                     of a mirror level 1 page table.
1124 * @table: Target level 1 page table.
1125 * @idx: Index of the entry to access.
1126 *
1127 * Technically this function returns a pointer to a slot in a raw level 1 page
1128 * table, since the returned "entry" is not guaranteed to be valid. The caller
1129 * must verify the validity of the entry at the returned address (perhaps using
1130 * pvr_page_table_l1_entry_raw_is_valid()) before reading or overwriting it.
1131 *
1132 * The value of @idx is not checked here; it is the callers responsibility to
1133 * ensure @idx refers to a valid index within @table before dereferencing the
1134 * returned pointer.
1135 *
1136 * Return:
1137 * A pointer to the requested raw level 1 page table entry.
1138 */
1139static struct pvr_page_table_l1_entry_raw *
1140pvr_page_table_l1_get_entry_raw(struct pvr_page_table_l1 *table, u16 idx)
1141{
1142	return &pvr_page_table_l1_get_raw(table)->entries[idx];
1143}
1144
1145/**
1146 * pvr_page_table_l1_entry_is_valid() - Check if a level 1 page table entry is
1147 *                                      marked as valid.
1148 * @table: Target level 1 page table.
1149 * @idx: Index of the entry to check.
1150 *
1151 * The value of @idx is not checked here; it is the callers responsibility to
1152 * ensure @idx refers to a valid index within @table before calling this
1153 * function.
1154 */
1155static bool
1156pvr_page_table_l1_entry_is_valid(struct pvr_page_table_l1 *table, u16 idx)
1157{
1158	struct pvr_page_table_l1_entry_raw entry_raw =
1159		*pvr_page_table_l1_get_entry_raw(table, idx);
1160
1161	return pvr_page_table_l1_entry_raw_is_valid(entry_raw);
1162}
1163
1164/**
1165 * struct pvr_page_table_l0 - A wrapped level 0 page table.
1166 *
1167 * To access the raw part of this table, use pvr_page_table_l0_get_raw().
1168 * Alternatively to access a raw entry directly, use
1169 * pvr_page_table_l0_get_entry_raw().
1170 *
1171 * There is no mirror representation of an individual page, so this type has no
1172 * &entries member.
1173 */
1174struct pvr_page_table_l0 {
1175	/**
1176	 * @backing_page: A handle to the memory which holds the raw
1177	 * equivalent of this table. **For internal use only.**
1178	 */
1179	struct pvr_mmu_backing_page backing_page;
1180
1181	union {
1182		/**
1183		 * @parent: The parent of this node in the page table tree structure.
1184		 *
1185		 * This is also a mirror table.
1186		 *
1187		 * Only valid when the L0 page table is active. When the L0 page table
1188		 * has been removed and queued for destruction, the next_free field
1189		 * should be used instead.
1190		 */
1191		struct pvr_page_table_l1 *parent;
1192
1193		/**
1194		 * @next_free: Pointer to the next L0 page table to take/free.
1195		 *
1196		 * Used to form a linked list of L0 page tables. This is used
1197		 * when preallocating tables and when the page table has been
1198		 * removed and queued for destruction.
1199		 */
1200		struct pvr_page_table_l0 *next_free;
1201	};
1202
1203	/**
1204	 * @parent_idx: The index of the entry in the parent table (see
1205	 * @parent) which corresponds to this table.
1206	 */
1207	u16 parent_idx;
1208
1209	/**
1210	 * @entry_count: The current number of valid entries (that we know of)
1211	 * in this table. This value is essentially a refcount - the table is
1212	 * destroyed when this value is decremented to zero by
1213	 * pvr_page_table_l0_remove().
1214	 */
1215	u16 entry_count;
1216};
1217
1218/**
1219 * pvr_page_table_l0_init() - Initialize a level 0 page table.
1220 * @table: Target level 0 page table.
1221 * @pvr_dev: Target PowerVR device
1222 *
1223 * When this function returns successfully, @table is still not considered
1224 * valid. It must be inserted into the page table tree structure with
1225 * pvr_page_table_l1_insert() before it is ready for use.
1226 *
1227 * It is expected that @table be zeroed (e.g. from kzalloc()) before calling
1228 * this function.
1229 *
1230 * Return:
1231 *  * 0 on success, or
1232 *  * Any error encountered while intializing &table->backing_page using
1233 *    pvr_mmu_backing_page_init().
1234 */
1235static int
1236pvr_page_table_l0_init(struct pvr_page_table_l0 *table,
1237		       struct pvr_device *pvr_dev)
1238{
1239	table->parent_idx = PVR_IDX_INVALID;
1240
1241	return pvr_mmu_backing_page_init(&table->backing_page, pvr_dev);
1242}
1243
1244/**
1245 * pvr_page_table_l0_free() - Teardown a level 0 page table.
1246 * @table: Target level 0 page table.
1247 *
1248 * It is an error to attempt to use @table after calling this function, even
1249 * indirectly. This includes calling pvr_page_table_l1_remove(), which must
1250 * be called *before* pvr_page_table_l0_free().
1251 */
1252static void
1253pvr_page_table_l0_free(struct pvr_page_table_l0 *table)
1254{
1255	pvr_mmu_backing_page_fini(&table->backing_page);
1256	kfree(table);
1257}
1258
1259/**
1260 * pvr_page_table_l0_sync() - Flush a level 0 page table from the CPU to the
1261 *                            device.
1262 * @table: Target level 0 page table.
1263 *
1264 * This is just a thin wrapper around pvr_mmu_backing_page_sync(), so the
1265 * warning there applies here too: **Only call pvr_page_table_l0_sync() once
1266 * you're sure you have no more changes to make to** @table **in the immediate
1267 * future.**
1268 *
1269 * If child pages of @table also need to be flushed, this should be done first
1270 * using a DMA sync function (e.g. dma_sync_sg_for_device()) *before* calling
1271 * this function.
1272 */
1273static void
1274pvr_page_table_l0_sync(struct pvr_page_table_l0 *table)
1275{
1276	pvr_mmu_backing_page_sync(&table->backing_page, PVR_MMU_SYNC_LEVEL_0_FLAGS);
1277}
1278
1279/**
1280 * pvr_page_table_l0_get_raw() - Access the raw equivalent of a mirror level 0
1281 *                               page table.
1282 * @table: Target level 0 page table.
1283 *
1284 * Essentially returns the CPU address of the raw equivalent of @table, cast to
1285 * a &struct pvr_page_table_l0_raw pointer.
1286 *
1287 * You probably want to call pvr_page_table_l0_get_entry_raw() instead.
1288 *
1289 * Return:
1290 * The raw equivalent of @table.
1291 */
1292static struct pvr_page_table_l0_raw *
1293pvr_page_table_l0_get_raw(struct pvr_page_table_l0 *table)
1294{
1295	return table->backing_page.host_ptr;
1296}
1297
1298/**
1299 * pvr_page_table_l0_get_entry_raw() - Access an entry from the raw equivalent
1300 *                                     of a mirror level 0 page table.
1301 * @table: Target level 0 page table.
1302 * @idx: Index of the entry to access.
1303 *
1304 * Technically this function returns a pointer to a slot in a raw level 0 page
1305 * table, since the returned "entry" is not guaranteed to be valid. The caller
1306 * must verify the validity of the entry at the returned address (perhaps using
1307 * pvr_page_table_l0_entry_raw_is_valid()) before reading or overwriting it.
1308 *
1309 * The value of @idx is not checked here; it is the callers responsibility to
1310 * ensure @idx refers to a valid index within @table before dereferencing the
1311 * returned pointer. This is espcially important for level 0 page tables, which
1312 * can have a variable number of entries.
1313 *
1314 * Return:
1315 * A pointer to the requested raw level 0 page table entry.
1316 */
1317static struct pvr_page_table_l0_entry_raw *
1318pvr_page_table_l0_get_entry_raw(struct pvr_page_table_l0 *table, u16 idx)
1319{
1320	return &pvr_page_table_l0_get_raw(table)->entries[idx];
1321}
1322
1323/**
1324 * pvr_page_table_l0_entry_is_valid() - Check if a level 0 page table entry is
1325 *                                      marked as valid.
1326 * @table: Target level 0 page table.
1327 * @idx: Index of the entry to check.
1328 *
1329 * The value of @idx is not checked here; it is the callers responsibility to
1330 * ensure @idx refers to a valid index within @table before calling this
1331 * function.
1332 */
1333static bool
1334pvr_page_table_l0_entry_is_valid(struct pvr_page_table_l0 *table, u16 idx)
1335{
1336	struct pvr_page_table_l0_entry_raw entry_raw =
1337		*pvr_page_table_l0_get_entry_raw(table, idx);
1338
1339	return pvr_page_table_l0_entry_raw_is_valid(entry_raw);
1340}
1341
1342/**
1343 * struct pvr_mmu_context - context holding data for operations at page
1344 * catalogue level, intended for use with a VM context.
1345 */
1346struct pvr_mmu_context {
1347	/** @pvr_dev: The PVR device associated with the owning VM context. */
1348	struct pvr_device *pvr_dev;
1349
1350	/** @page_table_l2: The MMU table root. */
1351	struct pvr_page_table_l2 page_table_l2;
1352};
1353
1354/**
1355 * struct pvr_page_table_ptr - A reference to a single physical page as indexed
1356 * by the page table structure.
1357 *
1358 * Intended for embedding in a &struct pvr_mmu_op_context.
1359 */
1360struct pvr_page_table_ptr {
1361	/**
1362	 * @l1_table: A cached handle to the level 1 page table the
1363	 * context is currently traversing.
1364	 */
1365	struct pvr_page_table_l1 *l1_table;
1366
1367	/**
1368	 * @l0_table: A cached handle to the level 0 page table the
1369	 * context is currently traversing.
1370	 */
1371	struct pvr_page_table_l0 *l0_table;
1372
1373	/**
1374	 * @l2_idx: Index into the level 2 page table the context is
1375	 * currently referencing.
1376	 */
1377	u16 l2_idx;
1378
1379	/**
1380	 * @l1_idx: Index into the level 1 page table the context is
1381	 * currently referencing.
1382	 */
1383	u16 l1_idx;
1384
1385	/**
1386	 * @l0_idx: Index into the level 0 page table the context is
1387	 * currently referencing.
1388	 */
1389	u16 l0_idx;
1390};
1391
1392/**
1393 * struct pvr_mmu_op_context - context holding data for individual
1394 * device-virtual mapping operations. Intended for use with a VM bind operation.
1395 */
1396struct pvr_mmu_op_context {
1397	/** @mmu_ctx: The MMU context associated with the owning VM context. */
1398	struct pvr_mmu_context *mmu_ctx;
1399
1400	/** @map: Data specifically for map operations. */
1401	struct {
1402		/**
1403		 * @sgt: Scatter gather table containing pages pinned for use by
1404		 * this context - these are currently pinned when initialising
1405		 * the VM bind operation.
1406		 */
1407		struct sg_table *sgt;
1408
1409		/** @sgt_offset: Start address of the device-virtual mapping. */
1410		u64 sgt_offset;
1411
1412		/**
1413		 * @l1_prealloc_tables: Preallocated l1 page table objects
1414		 * use by this context when creating a page mapping. Linked list
1415		 * fully created during initialisation.
1416		 */
1417		struct pvr_page_table_l1 *l1_prealloc_tables;
1418
1419		/**
1420		 * @l0_prealloc_tables: Preallocated l0 page table objects
1421		 * use by this context when creating a page mapping. Linked list
1422		 * fully created during initialisation.
1423		 */
1424		struct pvr_page_table_l0 *l0_prealloc_tables;
1425	} map;
1426
1427	/** @unmap: Data specifically for unmap operations. */
1428	struct {
1429		/**
1430		 * @l1_free_tables: Collects page table objects freed by unmap
1431		 * ops. Linked list empty at creation.
1432		 */
1433		struct pvr_page_table_l1 *l1_free_tables;
1434
1435		/**
1436		 * @l0_free_tables: Collects page table objects freed by unmap
1437		 * ops. Linked list empty at creation.
1438		 */
1439		struct pvr_page_table_l0 *l0_free_tables;
1440	} unmap;
1441
1442	/**
1443	 * @curr_page: A reference to a single physical page as indexed by the
1444	 * page table structure.
1445	 */
1446	struct pvr_page_table_ptr curr_page;
1447
1448	/**
1449	 * @sync_level_required: The maximum level of the page table tree
1450	 * structure which has (possibly) been modified since it was last
1451	 * flushed to the device.
1452	 *
1453	 * This field should only be set with pvr_mmu_op_context_require_sync()
1454	 * or indirectly by pvr_mmu_op_context_sync_partial().
1455	 */
1456	enum pvr_mmu_sync_level sync_level_required;
1457};
1458
1459/**
1460 * pvr_page_table_l2_insert() - Insert an entry referring to a level 1 page
1461 * table into a level 2 page table.
1462 * @op_ctx: Target MMU op context pointing at the entry to insert the L1 page
1463 * table into.
1464 * @child_table: Target level 1 page table to be referenced by the new entry.
1465 *
1466 * It is the caller's responsibility to ensure @op_ctx.curr_page points to a
1467 * valid L2 entry.
1468 *
1469 * It is the caller's responsibility to execute any memory barries to ensure
1470 * that the creation of @child_table is ordered before the L2 entry is inserted.
1471 */
1472static void
1473pvr_page_table_l2_insert(struct pvr_mmu_op_context *op_ctx,
1474			 struct pvr_page_table_l1 *child_table)
1475{
1476	struct pvr_page_table_l2 *l2_table =
1477		&op_ctx->mmu_ctx->page_table_l2;
1478	struct pvr_page_table_l2_entry_raw *entry_raw =
1479		pvr_page_table_l2_get_entry_raw(l2_table,
1480						op_ctx->curr_page.l2_idx);
1481
1482	pvr_page_table_l2_entry_raw_set(entry_raw,
1483					child_table->backing_page.dma_addr);
1484
1485	child_table->parent = l2_table;
1486	child_table->parent_idx = op_ctx->curr_page.l2_idx;
1487	l2_table->entries[op_ctx->curr_page.l2_idx] = child_table;
1488	++l2_table->entry_count;
1489	op_ctx->curr_page.l1_table = child_table;
1490}
1491
1492/**
1493 * pvr_page_table_l2_remove() - Remove a level 1 page table from a level 2 page
1494 * table.
1495 * @op_ctx: Target MMU op context pointing at the L2 entry to remove.
1496 *
1497 * It is the caller's responsibility to ensure @op_ctx.curr_page points to a
1498 * valid L2 entry.
1499 */
1500static void
1501pvr_page_table_l2_remove(struct pvr_mmu_op_context *op_ctx)
1502{
1503	struct pvr_page_table_l2 *l2_table =
1504		&op_ctx->mmu_ctx->page_table_l2;
1505	struct pvr_page_table_l2_entry_raw *entry_raw =
1506		pvr_page_table_l2_get_entry_raw(l2_table,
1507						op_ctx->curr_page.l1_table->parent_idx);
1508
1509	WARN_ON(op_ctx->curr_page.l1_table->parent != l2_table);
1510
1511	pvr_page_table_l2_entry_raw_clear(entry_raw);
1512
1513	l2_table->entries[op_ctx->curr_page.l1_table->parent_idx] = NULL;
1514	op_ctx->curr_page.l1_table->parent_idx = PVR_IDX_INVALID;
1515	op_ctx->curr_page.l1_table->next_free = op_ctx->unmap.l1_free_tables;
1516	op_ctx->unmap.l1_free_tables = op_ctx->curr_page.l1_table;
1517	op_ctx->curr_page.l1_table = NULL;
1518
1519	--l2_table->entry_count;
1520}
1521
1522/**
1523 * pvr_page_table_l1_insert() - Insert an entry referring to a level 0 page
1524 * table into a level 1 page table.
1525 * @op_ctx: Target MMU op context pointing at the entry to insert the L0 page
1526 * table into.
1527 * @child_table: L0 page table to insert.
1528 *
1529 * It is the caller's responsibility to ensure @op_ctx.curr_page points to a
1530 * valid L1 entry.
1531 *
1532 * It is the caller's responsibility to execute any memory barries to ensure
1533 * that the creation of @child_table is ordered before the L1 entry is inserted.
1534 */
1535static void
1536pvr_page_table_l1_insert(struct pvr_mmu_op_context *op_ctx,
1537			 struct pvr_page_table_l0 *child_table)
1538{
1539	struct pvr_page_table_l1_entry_raw *entry_raw =
1540		pvr_page_table_l1_get_entry_raw(op_ctx->curr_page.l1_table,
1541						op_ctx->curr_page.l1_idx);
1542
1543	pvr_page_table_l1_entry_raw_set(entry_raw,
1544					child_table->backing_page.dma_addr);
1545
1546	child_table->parent = op_ctx->curr_page.l1_table;
1547	child_table->parent_idx = op_ctx->curr_page.l1_idx;
1548	op_ctx->curr_page.l1_table->entries[op_ctx->curr_page.l1_idx] = child_table;
1549	++op_ctx->curr_page.l1_table->entry_count;
1550	op_ctx->curr_page.l0_table = child_table;
1551}
1552
1553/**
1554 * pvr_page_table_l1_remove() - Remove a level 0 page table from a level 1 page
1555 *                              table.
1556 * @op_ctx: Target MMU op context pointing at the L1 entry to remove.
1557 *
1558 * If this function results in the L1 table becoming empty, it will be removed
1559 * from its parent level 2 page table and destroyed.
1560 *
1561 * It is the caller's responsibility to ensure @op_ctx.curr_page points to a
1562 * valid L1 entry.
1563 */
1564static void
1565pvr_page_table_l1_remove(struct pvr_mmu_op_context *op_ctx)
1566{
1567	struct pvr_page_table_l1_entry_raw *entry_raw =
1568		pvr_page_table_l1_get_entry_raw(op_ctx->curr_page.l0_table->parent,
1569						op_ctx->curr_page.l0_table->parent_idx);
1570
1571	WARN_ON(op_ctx->curr_page.l0_table->parent !=
1572		op_ctx->curr_page.l1_table);
1573
1574	pvr_page_table_l1_entry_raw_clear(entry_raw);
1575
1576	op_ctx->curr_page.l1_table->entries[op_ctx->curr_page.l0_table->parent_idx] = NULL;
1577	op_ctx->curr_page.l0_table->parent_idx = PVR_IDX_INVALID;
1578	op_ctx->curr_page.l0_table->next_free = op_ctx->unmap.l0_free_tables;
1579	op_ctx->unmap.l0_free_tables = op_ctx->curr_page.l0_table;
1580	op_ctx->curr_page.l0_table = NULL;
1581
1582	if (--op_ctx->curr_page.l1_table->entry_count == 0) {
1583		/* Clear the parent L2 page table entry. */
1584		if (op_ctx->curr_page.l1_table->parent_idx != PVR_IDX_INVALID)
1585			pvr_page_table_l2_remove(op_ctx);
1586	}
1587}
1588
1589/**
1590 * pvr_page_table_l0_insert() - Insert an entry referring to a physical page
1591 * into a level 0 page table.
1592 * @op_ctx: Target MMU op context pointing at the L0 entry to insert.
1593 * @dma_addr: Target DMA address to be referenced by the new entry.
1594 * @flags: Page options to be stored in the new entry.
1595 *
1596 * It is the caller's responsibility to ensure @op_ctx.curr_page points to a
1597 * valid L0 entry.
1598 */
1599static void
1600pvr_page_table_l0_insert(struct pvr_mmu_op_context *op_ctx,
1601			 dma_addr_t dma_addr, struct pvr_page_flags_raw flags)
1602{
1603	struct pvr_page_table_l0_entry_raw *entry_raw =
1604		pvr_page_table_l0_get_entry_raw(op_ctx->curr_page.l0_table,
1605						op_ctx->curr_page.l0_idx);
1606
1607	pvr_page_table_l0_entry_raw_set(entry_raw, dma_addr, flags);
1608
1609	/*
1610	 * There is no entry to set here - we don't keep a mirror of
1611	 * individual pages.
1612	 */
1613
1614	++op_ctx->curr_page.l0_table->entry_count;
1615}
1616
1617/**
1618 * pvr_page_table_l0_remove() - Remove a physical page from a level 0 page
1619 * table.
1620 * @op_ctx: Target MMU op context pointing at the L0 entry to remove.
1621 *
1622 * If this function results in the L0 table becoming empty, it will be removed
1623 * from its parent L1 page table and destroyed.
1624 *
1625 * It is the caller's responsibility to ensure @op_ctx.curr_page points to a
1626 * valid L0 entry.
1627 */
1628static void
1629pvr_page_table_l0_remove(struct pvr_mmu_op_context *op_ctx)
1630{
1631	struct pvr_page_table_l0_entry_raw *entry_raw =
1632		pvr_page_table_l0_get_entry_raw(op_ctx->curr_page.l0_table,
1633						op_ctx->curr_page.l0_idx);
1634
1635	pvr_page_table_l0_entry_raw_clear(entry_raw);
1636
1637	/*
1638	 * There is no entry to clear here - we don't keep a mirror of
1639	 * individual pages.
1640	 */
1641
1642	if (--op_ctx->curr_page.l0_table->entry_count == 0) {
1643		/* Clear the parent L1 page table entry. */
1644		if (op_ctx->curr_page.l0_table->parent_idx != PVR_IDX_INVALID)
1645			pvr_page_table_l1_remove(op_ctx);
1646	}
1647}
1648
1649/**
1650 * DOC: Page table index utilities
1651 */
1652
1653/**
1654 * pvr_page_table_l2_idx() - Calculate the level 2 page table index for a
1655 *                           device-virtual address.
1656 * @device_addr: Target device-virtual address.
1657 *
1658 * This function does not perform any bounds checking - it is the caller's
1659 * responsibility to ensure that @device_addr is valid before interpreting
1660 * the result.
1661 *
1662 * Return:
1663 * The index into a level 2 page table corresponding to @device_addr.
1664 */
1665static u16
1666pvr_page_table_l2_idx(u64 device_addr)
1667{
1668	return (device_addr & ~ROGUE_MMUCTRL_VADDR_PC_INDEX_CLRMSK) >>
1669	       ROGUE_MMUCTRL_VADDR_PC_INDEX_SHIFT;
1670}
1671
1672/**
1673 * pvr_page_table_l1_idx() - Calculate the level 1 page table index for a
1674 *                           device-virtual address.
1675 * @device_addr: Target device-virtual address.
1676 *
1677 * This function does not perform any bounds checking - it is the caller's
1678 * responsibility to ensure that @device_addr is valid before interpreting
1679 * the result.
1680 *
1681 * Return:
1682 * The index into a level 1 page table corresponding to @device_addr.
1683 */
1684static u16
1685pvr_page_table_l1_idx(u64 device_addr)
1686{
1687	return (device_addr & ~ROGUE_MMUCTRL_VADDR_PD_INDEX_CLRMSK) >>
1688	       ROGUE_MMUCTRL_VADDR_PD_INDEX_SHIFT;
1689}
1690
1691/**
1692 * pvr_page_table_l0_idx() - Calculate the level 0 page table index for a
1693 *                           device-virtual address.
1694 * @device_addr: Target device-virtual address.
1695 *
1696 * This function does not perform any bounds checking - it is the caller's
1697 * responsibility to ensure that @device_addr is valid before interpreting
1698 * the result.
1699 *
1700 * Return:
1701 * The index into a level 0 page table corresponding to @device_addr.
1702 */
1703static u16
1704pvr_page_table_l0_idx(u64 device_addr)
1705{
1706	return (device_addr & ~ROGUE_MMUCTRL_VADDR_PT_INDEX_CLRMSK) >>
1707	       ROGUE_MMUCTRL_PAGE_X_RANGE_SHIFT;
1708}
1709
1710/**
1711 * DOC: High-level page table operations
1712 */
1713
1714/**
1715 * pvr_page_table_l1_get_or_insert() - Retrieves (optionally inserting if
1716 * necessary) a level 1 page table from the specified level 2 page table entry.
1717 * @op_ctx: Target MMU op context.
1718 * @should_insert: [IN] Specifies whether new page tables should be inserted
1719 * when empty page table entries are encountered during traversal.
1720 *
1721 * Return:
1722 *  * 0 on success, or
1723 *
1724 *    If @should_insert is %false:
1725 *     * -%ENXIO if a level 1 page table would have been inserted.
1726 *
1727 *    If @should_insert is %true:
1728 *     * Any error encountered while inserting the level 1 page table.
1729 */
1730static int
1731pvr_page_table_l1_get_or_insert(struct pvr_mmu_op_context *op_ctx,
1732				bool should_insert)
1733{
1734	struct pvr_page_table_l2 *l2_table =
1735		&op_ctx->mmu_ctx->page_table_l2;
1736	struct pvr_page_table_l1 *table;
1737
1738	if (pvr_page_table_l2_entry_is_valid(l2_table,
1739					     op_ctx->curr_page.l2_idx)) {
1740		op_ctx->curr_page.l1_table =
1741			l2_table->entries[op_ctx->curr_page.l2_idx];
1742		return 0;
1743	}
1744
1745	if (!should_insert)
1746		return -ENXIO;
1747
1748	/* Take a prealloced table. */
1749	table = op_ctx->map.l1_prealloc_tables;
1750	if (!table)
1751		return -ENOMEM;
1752
1753	/* Pop */
1754	op_ctx->map.l1_prealloc_tables = table->next_free;
1755	table->next_free = NULL;
1756
1757	/* Ensure new table is fully written out before adding to L2 page table. */
1758	wmb();
1759
1760	pvr_page_table_l2_insert(op_ctx, table);
1761
1762	return 0;
1763}
1764
1765/**
1766 * pvr_page_table_l0_get_or_insert() - Retrieves (optionally inserting if
1767 * necessary) a level 0 page table from the specified level 1 page table entry.
1768 * @op_ctx: Target MMU op context.
1769 * @should_insert: [IN] Specifies whether new page tables should be inserted
1770 * when empty page table entries are encountered during traversal.
1771 *
1772 * Return:
1773 *  * 0 on success,
1774 *
1775 *    If @should_insert is %false:
1776 *     * -%ENXIO if a level 0 page table would have been inserted.
1777 *
1778 *    If @should_insert is %true:
1779 *     * Any error encountered while inserting the level 0 page table.
1780 */
1781static int
1782pvr_page_table_l0_get_or_insert(struct pvr_mmu_op_context *op_ctx,
1783				bool should_insert)
1784{
1785	struct pvr_page_table_l0 *table;
1786
1787	if (pvr_page_table_l1_entry_is_valid(op_ctx->curr_page.l1_table,
1788					     op_ctx->curr_page.l1_idx)) {
1789		op_ctx->curr_page.l0_table =
1790			op_ctx->curr_page.l1_table->entries[op_ctx->curr_page.l1_idx];
1791		return 0;
1792	}
1793
1794	if (!should_insert)
1795		return -ENXIO;
1796
1797	/* Take a prealloced table. */
1798	table = op_ctx->map.l0_prealloc_tables;
1799	if (!table)
1800		return -ENOMEM;
1801
1802	/* Pop */
1803	op_ctx->map.l0_prealloc_tables = table->next_free;
1804	table->next_free = NULL;
1805
1806	/* Ensure new table is fully written out before adding to L1 page table. */
1807	wmb();
1808
1809	pvr_page_table_l1_insert(op_ctx, table);
1810
1811	return 0;
1812}
1813
1814/**
1815 * pvr_mmu_context_create() - Create an MMU context.
1816 * @pvr_dev: PVR device associated with owning VM context.
1817 *
1818 * Returns:
1819 *  * Newly created MMU context object on success, or
1820 *  * -%ENOMEM if no memory is available,
1821 *  * Any error code returned by pvr_page_table_l2_init().
1822 */
1823struct pvr_mmu_context *pvr_mmu_context_create(struct pvr_device *pvr_dev)
1824{
1825	struct pvr_mmu_context *ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1826	int err;
1827
1828	if (!ctx)
1829		return ERR_PTR(-ENOMEM);
1830
1831	err = pvr_page_table_l2_init(&ctx->page_table_l2, pvr_dev);
1832	if (err)
1833		return ERR_PTR(err);
1834
1835	ctx->pvr_dev = pvr_dev;
1836
1837	return ctx;
1838}
1839
1840/**
1841 * pvr_mmu_context_destroy() - Destroy an MMU context.
1842 * @ctx: Target MMU context.
1843 */
1844void pvr_mmu_context_destroy(struct pvr_mmu_context *ctx)
1845{
1846	pvr_page_table_l2_fini(&ctx->page_table_l2);
1847	kfree(ctx);
1848}
1849
1850/**
1851 * pvr_mmu_get_root_table_dma_addr() - Get the DMA address of the root of the
1852 * page table structure behind a VM context.
1853 * @ctx: Target MMU context.
1854 */
1855dma_addr_t pvr_mmu_get_root_table_dma_addr(struct pvr_mmu_context *ctx)
1856{
1857	return ctx->page_table_l2.backing_page.dma_addr;
1858}
1859
1860/**
1861 * pvr_page_table_l1_alloc() - Allocate a l1 page_table object.
1862 * @ctx: MMU context of owning VM context.
1863 *
1864 * Returns:
1865 *  * Newly created page table object on success, or
1866 *  * -%ENOMEM if no memory is available,
1867 *  * Any error code returned by pvr_page_table_l1_init().
1868 */
1869static struct pvr_page_table_l1 *
1870pvr_page_table_l1_alloc(struct pvr_mmu_context *ctx)
1871{
1872	int err;
1873
1874	struct pvr_page_table_l1 *table =
1875		kzalloc(sizeof(*table), GFP_KERNEL);
1876
1877	if (!table)
1878		return ERR_PTR(-ENOMEM);
1879
1880	err = pvr_page_table_l1_init(table, ctx->pvr_dev);
1881	if (err) {
1882		kfree(table);
1883		return ERR_PTR(err);
1884	}
1885
1886	return table;
1887}
1888
1889/**
1890 * pvr_page_table_l0_alloc() - Allocate a l0 page_table object.
1891 * @ctx: MMU context of owning VM context.
1892 *
1893 * Returns:
1894 *  * Newly created page table object on success, or
1895 *  * -%ENOMEM if no memory is available,
1896 *  * Any error code returned by pvr_page_table_l0_init().
1897 */
1898static struct pvr_page_table_l0 *
1899pvr_page_table_l0_alloc(struct pvr_mmu_context *ctx)
1900{
1901	int err;
1902
1903	struct pvr_page_table_l0 *table =
1904		kzalloc(sizeof(*table), GFP_KERNEL);
1905
1906	if (!table)
1907		return ERR_PTR(-ENOMEM);
1908
1909	err = pvr_page_table_l0_init(table, ctx->pvr_dev);
1910	if (err) {
1911		kfree(table);
1912		return ERR_PTR(err);
1913	}
1914
1915	return table;
1916}
1917
1918/**
1919 * pvr_mmu_op_context_require_sync() - Mark an MMU op context as requiring a
1920 * sync operation for the referenced page tables up to a specified level.
1921 * @op_ctx: Target MMU op context.
1922 * @level: Maximum page table level for which a sync is required.
1923 */
1924static void
1925pvr_mmu_op_context_require_sync(struct pvr_mmu_op_context *op_ctx,
1926				enum pvr_mmu_sync_level level)
1927{
1928	if (op_ctx->sync_level_required < level)
1929		op_ctx->sync_level_required = level;
1930}
1931
1932/**
1933 * pvr_mmu_op_context_sync_manual() - Trigger a sync of some or all of the
1934 * page tables referenced by a MMU op context.
1935 * @op_ctx: Target MMU op context.
1936 * @level: Maximum page table level to sync.
1937 *
1938 * Do not call this function directly. Instead use
1939 * pvr_mmu_op_context_sync_partial() which is checked against the current
1940 * value of &op_ctx->sync_level_required as set by
1941 * pvr_mmu_op_context_require_sync().
1942 */
1943static void
1944pvr_mmu_op_context_sync_manual(struct pvr_mmu_op_context *op_ctx,
1945			       enum pvr_mmu_sync_level level)
1946{
1947	/*
1948	 * We sync the page table levels in ascending order (starting from the
1949	 * leaf node) to ensure consistency.
1950	 */
1951
1952	WARN_ON(level < PVR_MMU_SYNC_LEVEL_NONE);
1953
1954	if (level <= PVR_MMU_SYNC_LEVEL_NONE)
1955		return;
1956
1957	if (op_ctx->curr_page.l0_table)
1958		pvr_page_table_l0_sync(op_ctx->curr_page.l0_table);
1959
1960	if (level < PVR_MMU_SYNC_LEVEL_1)
1961		return;
1962
1963	if (op_ctx->curr_page.l1_table)
1964		pvr_page_table_l1_sync(op_ctx->curr_page.l1_table);
1965
1966	if (level < PVR_MMU_SYNC_LEVEL_2)
1967		return;
1968
1969	pvr_page_table_l2_sync(&op_ctx->mmu_ctx->page_table_l2);
1970}
1971
1972/**
1973 * pvr_mmu_op_context_sync_partial() - Trigger a sync of some or all of the
1974 * page tables referenced by a MMU op context.
1975 * @op_ctx: Target MMU op context.
1976 * @level: Requested page table level to sync up to (inclusive).
1977 *
1978 * If @level is greater than the maximum level recorded by @op_ctx as requiring
1979 * a sync operation, only the previously recorded maximum will be used.
1980 *
1981 * Additionally, if @level is greater than or equal to the maximum level
1982 * recorded by @op_ctx as requiring a sync operation, that maximum level will be
1983 * reset as a full sync will be performed. This is equivalent to calling
1984 * pvr_mmu_op_context_sync().
1985 */
1986static void
1987pvr_mmu_op_context_sync_partial(struct pvr_mmu_op_context *op_ctx,
1988				enum pvr_mmu_sync_level level)
1989{
1990	/*
1991	 * If the requested sync level is greater than or equal to the
1992	 * currently required sync level, we do two things:
1993	 *  * Don't waste time syncing levels we haven't previously marked as
1994	 *    requiring a sync, and
1995	 *  * Reset the required sync level since we are about to sync
1996	 *    everything that was previously marked as requiring a sync.
1997	 */
1998	if (level >= op_ctx->sync_level_required) {
1999		level = op_ctx->sync_level_required;
2000		op_ctx->sync_level_required = PVR_MMU_SYNC_LEVEL_NONE;
2001	}
2002
2003	pvr_mmu_op_context_sync_manual(op_ctx, level);
2004}
2005
2006/**
2007 * pvr_mmu_op_context_sync() - Trigger a sync of every page table referenced by
2008 * a MMU op context.
2009 * @op_ctx: Target MMU op context.
2010 *
2011 * The maximum level marked internally as requiring a sync will be reset so
2012 * that subsequent calls to this function will be no-ops unless @op_ctx is
2013 * otherwise updated.
2014 */
2015static void
2016pvr_mmu_op_context_sync(struct pvr_mmu_op_context *op_ctx)
2017{
2018	pvr_mmu_op_context_sync_manual(op_ctx, op_ctx->sync_level_required);
2019
2020	op_ctx->sync_level_required = PVR_MMU_SYNC_LEVEL_NONE;
2021}
2022
2023/**
2024 * pvr_mmu_op_context_load_tables() - Load pointers to tables in each level of
2025 * the page table tree structure needed to reference the physical page
2026 * referenced by a MMU op context.
2027 * @op_ctx: Target MMU op context.
2028 * @should_create: Specifies whether new page tables should be created when
2029 * empty page table entries are encountered during traversal.
2030 * @load_level_required: Maximum page table level to load.
2031 *
2032 * If @should_create is %true, this function may modify the stored required
2033 * sync level of @op_ctx as new page tables are created and inserted into their
2034 * respective parents.
2035 *
2036 * Since there is only one root page table, it is technically incorrect to call
2037 * this function with a value of @load_level_required greater than or equal to
2038 * the root level number. However, this is not explicitly disallowed here.
2039 *
2040 * Return:
2041 *  * 0 on success,
2042 *  * Any error returned by pvr_page_table_l1_get_or_create() if
2043 *    @load_level_required >= 1 except -%ENXIO, or
2044 *  * Any error returned by pvr_page_table_l0_get_or_create() if
2045 *    @load_level_required >= 0 except -%ENXIO.
2046 */
2047static int
2048pvr_mmu_op_context_load_tables(struct pvr_mmu_op_context *op_ctx,
2049			       bool should_create,
2050			       enum pvr_mmu_sync_level load_level_required)
2051{
2052	const struct pvr_page_table_l1 *l1_head_before =
2053		op_ctx->map.l1_prealloc_tables;
2054	const struct pvr_page_table_l0 *l0_head_before =
2055		op_ctx->map.l0_prealloc_tables;
2056	int err;
2057
2058	/* Clear tables we're about to fetch in case of error states. */
2059	if (load_level_required >= PVR_MMU_SYNC_LEVEL_1)
2060		op_ctx->curr_page.l1_table = NULL;
2061
2062	if (load_level_required >= PVR_MMU_SYNC_LEVEL_0)
2063		op_ctx->curr_page.l0_table = NULL;
2064
2065	/* Get or create L1 page table. */
2066	if (load_level_required >= PVR_MMU_SYNC_LEVEL_1) {
2067		err = pvr_page_table_l1_get_or_insert(op_ctx, should_create);
2068		if (err) {
2069			/*
2070			 * If @should_create is %false and no L1 page table was
2071			 * found, return early but without an error. Since
2072			 * pvr_page_table_l1_get_or_create() can only return
2073			 * -%ENXIO if @should_create is %false, there is no
2074			 * need to check it here.
2075			 */
2076			if (err == -ENXIO)
2077				err = 0;
2078
2079			return err;
2080		}
2081	}
2082
2083	/* Get or create L0 page table. */
2084	if (load_level_required >= PVR_MMU_SYNC_LEVEL_0) {
2085		err = pvr_page_table_l0_get_or_insert(op_ctx, should_create);
2086		if (err) {
2087			/*
2088			 * If @should_create is %false and no L0 page table was
2089			 * found, return early but without an error. Since
2090			 * pvr_page_table_l0_get_or_insert() can only return
2091			 * -%ENXIO if @should_create is %false, there is no
2092			 * need to check it here.
2093			 */
2094			if (err == -ENXIO)
2095				err = 0;
2096
2097			/*
2098			 * At this point, an L1 page table could have been
2099			 * inserted but is now empty due to the failed attempt
2100			 * at inserting an L0 page table. In this instance, we
2101			 * must remove the empty L1 page table ourselves as
2102			 * pvr_page_table_l1_remove() is never called as part
2103			 * of the error path in
2104			 * pvr_page_table_l0_get_or_insert().
2105			 */
2106			if (l1_head_before != op_ctx->map.l1_prealloc_tables) {
2107				pvr_page_table_l2_remove(op_ctx);
2108				pvr_mmu_op_context_require_sync(op_ctx, PVR_MMU_SYNC_LEVEL_2);
2109			}
2110
2111			return err;
2112		}
2113	}
2114
2115	/*
2116	 * A sync is only needed if table objects were inserted. This can be
2117	 * inferred by checking if the pointer at the head of the linked list
2118	 * has changed.
2119	 */
2120	if (l1_head_before != op_ctx->map.l1_prealloc_tables)
2121		pvr_mmu_op_context_require_sync(op_ctx, PVR_MMU_SYNC_LEVEL_2);
2122	else if (l0_head_before != op_ctx->map.l0_prealloc_tables)
2123		pvr_mmu_op_context_require_sync(op_ctx, PVR_MMU_SYNC_LEVEL_1);
2124
2125	return 0;
2126}
2127
2128/**
2129 * pvr_mmu_op_context_set_curr_page() - Reassign the current page of an MMU op
2130 * context, syncing any page tables previously assigned to it which are no
2131 * longer relevant.
2132 * @op_ctx: Target MMU op context.
2133 * @device_addr: New pointer target.
2134 * @should_create: Specify whether new page tables should be created when
2135 * empty page table entries are encountered during traversal.
2136 *
2137 * This function performs a full sync on the pointer, regardless of which
2138 * levels are modified.
2139 *
2140 * Return:
2141 *  * 0 on success, or
2142 *  * Any error returned by pvr_mmu_op_context_load_tables().
2143 */
2144static int
2145pvr_mmu_op_context_set_curr_page(struct pvr_mmu_op_context *op_ctx,
2146				 u64 device_addr, bool should_create)
2147{
2148	pvr_mmu_op_context_sync(op_ctx);
2149
2150	op_ctx->curr_page.l2_idx = pvr_page_table_l2_idx(device_addr);
2151	op_ctx->curr_page.l1_idx = pvr_page_table_l1_idx(device_addr);
2152	op_ctx->curr_page.l0_idx = pvr_page_table_l0_idx(device_addr);
2153	op_ctx->curr_page.l1_table = NULL;
2154	op_ctx->curr_page.l0_table = NULL;
2155
2156	return pvr_mmu_op_context_load_tables(op_ctx, should_create,
2157					      PVR_MMU_SYNC_LEVEL_1);
2158}
2159
2160/**
2161 * pvr_mmu_op_context_next_page() - Advance the current page of an MMU op
2162 * context.
2163 * @op_ctx: Target MMU op context.
2164 * @should_create: Specify whether new page tables should be created when
2165 * empty page table entries are encountered during traversal.
2166 *
2167 * If @should_create is %false, it is the caller's responsibility to verify that
2168 * the state of the table references in @op_ctx is valid on return. If -%ENXIO
2169 * is returned, at least one of the table references is invalid. It should be
2170 * noted that @op_ctx as a whole will be left in a valid state if -%ENXIO is
2171 * returned, unlike other error codes. The caller should check which references
2172 * are invalid by comparing them to %NULL. Only &@ptr->l2_table is guaranteed
2173 * to be valid, since it represents the root of the page table tree structure.
2174 *
2175 * Return:
2176 *  * 0 on success,
2177 *  * -%EPERM if the operation would wrap at the top of the page table
2178 *    hierarchy,
2179 *  * -%ENXIO if @should_create is %false and a page table of any level would
2180 *    have otherwise been created, or
2181 *  * Any error returned while attempting to create missing page tables if
2182 *    @should_create is %true.
2183 */
2184static int
2185pvr_mmu_op_context_next_page(struct pvr_mmu_op_context *op_ctx,
2186			     bool should_create)
2187{
2188	s8 load_level_required = PVR_MMU_SYNC_LEVEL_NONE;
2189
2190	if (++op_ctx->curr_page.l0_idx != ROGUE_MMUCTRL_ENTRIES_PT_VALUE_X)
2191		goto load_tables;
2192
2193	op_ctx->curr_page.l0_idx = 0;
2194	load_level_required = PVR_MMU_SYNC_LEVEL_0;
2195
2196	if (++op_ctx->curr_page.l1_idx != ROGUE_MMUCTRL_ENTRIES_PD_VALUE)
2197		goto load_tables;
2198
2199	op_ctx->curr_page.l1_idx = 0;
2200	load_level_required = PVR_MMU_SYNC_LEVEL_1;
2201
2202	if (++op_ctx->curr_page.l2_idx != ROGUE_MMUCTRL_ENTRIES_PC_VALUE)
2203		goto load_tables;
2204
2205	/*
2206	 * If the pattern continued, we would set &op_ctx->curr_page.l2_idx to
2207	 * zero here. However, that would wrap the top layer of the page table
2208	 * hierarchy which is not a valid operation. Instead, we warn and return
2209	 * an error.
2210	 */
2211	WARN(true,
2212	     "%s(%p) attempted to loop the top of the page table hierarchy",
2213	     __func__, op_ctx);
2214	return -EPERM;
2215
2216	/* If indices have wrapped, we need to load new tables. */
2217load_tables:
2218	/* First, flush tables which will be unloaded. */
2219	pvr_mmu_op_context_sync_partial(op_ctx, load_level_required);
2220
2221	/* Then load tables from the required level down. */
2222	return pvr_mmu_op_context_load_tables(op_ctx, should_create,
2223					      load_level_required);
2224}
2225
2226/**
2227 * DOC: Single page operations
2228 */
2229
2230/**
2231 * pvr_page_create() - Create a device-virtual memory page and insert it into
2232 * a level 0 page table.
2233 * @op_ctx: Target MMU op context pointing at the device-virtual address of the
2234 * target page.
2235 * @dma_addr: DMA address of the physical page backing the created page.
2236 * @flags: Page options saved on the level 0 page table entry for reading by
2237 *         the device.
2238 *
2239 * Return:
2240 *  * 0 on success, or
2241 *  * -%EEXIST if the requested page already exists.
2242 */
2243static int
2244pvr_page_create(struct pvr_mmu_op_context *op_ctx, dma_addr_t dma_addr,
2245		struct pvr_page_flags_raw flags)
2246{
2247	/* Do not create a new page if one already exists. */
2248	if (pvr_page_table_l0_entry_is_valid(op_ctx->curr_page.l0_table,
2249					     op_ctx->curr_page.l0_idx)) {
2250		return -EEXIST;
2251	}
2252
2253	pvr_page_table_l0_insert(op_ctx, dma_addr, flags);
2254
2255	pvr_mmu_op_context_require_sync(op_ctx, PVR_MMU_SYNC_LEVEL_0);
2256
2257	return 0;
2258}
2259
2260/**
2261 * pvr_page_destroy() - Destroy a device page after removing it from its
2262 * parent level 0 page table.
2263 * @op_ctx: Target MMU op context.
2264 */
2265static void
2266pvr_page_destroy(struct pvr_mmu_op_context *op_ctx)
2267{
2268	/* Do nothing if the page does not exist. */
2269	if (!pvr_page_table_l0_entry_is_valid(op_ctx->curr_page.l0_table,
2270					      op_ctx->curr_page.l0_idx)) {
2271		return;
2272	}
2273
2274	/* Clear the parent L0 page table entry. */
2275	pvr_page_table_l0_remove(op_ctx);
2276
2277	pvr_mmu_op_context_require_sync(op_ctx, PVR_MMU_SYNC_LEVEL_0);
2278}
2279
2280/**
2281 * pvr_mmu_op_context_destroy() - Destroy an MMU op context.
2282 * @op_ctx: Target MMU op context.
2283 */
2284void pvr_mmu_op_context_destroy(struct pvr_mmu_op_context *op_ctx)
2285{
2286	const bool flush_caches =
2287		op_ctx->sync_level_required != PVR_MMU_SYNC_LEVEL_NONE;
2288
2289	pvr_mmu_op_context_sync(op_ctx);
2290
2291	/* Unmaps should be flushed immediately. Map flushes can be deferred. */
2292	if (flush_caches && !op_ctx->map.sgt)
2293		pvr_mmu_flush_exec(op_ctx->mmu_ctx->pvr_dev, true);
2294
2295	while (op_ctx->map.l0_prealloc_tables) {
2296		struct pvr_page_table_l0 *tmp = op_ctx->map.l0_prealloc_tables;
2297
2298		op_ctx->map.l0_prealloc_tables =
2299			op_ctx->map.l0_prealloc_tables->next_free;
2300		pvr_page_table_l0_free(tmp);
2301	}
2302
2303	while (op_ctx->map.l1_prealloc_tables) {
2304		struct pvr_page_table_l1 *tmp = op_ctx->map.l1_prealloc_tables;
2305
2306		op_ctx->map.l1_prealloc_tables =
2307			op_ctx->map.l1_prealloc_tables->next_free;
2308		pvr_page_table_l1_free(tmp);
2309	}
2310
2311	while (op_ctx->unmap.l0_free_tables) {
2312		struct pvr_page_table_l0 *tmp = op_ctx->unmap.l0_free_tables;
2313
2314		op_ctx->unmap.l0_free_tables =
2315			op_ctx->unmap.l0_free_tables->next_free;
2316		pvr_page_table_l0_free(tmp);
2317	}
2318
2319	while (op_ctx->unmap.l1_free_tables) {
2320		struct pvr_page_table_l1 *tmp = op_ctx->unmap.l1_free_tables;
2321
2322		op_ctx->unmap.l1_free_tables =
2323			op_ctx->unmap.l1_free_tables->next_free;
2324		pvr_page_table_l1_free(tmp);
2325	}
2326
2327	kfree(op_ctx);
2328}
2329
2330/**
2331 * pvr_mmu_op_context_create() - Create an MMU op context.
2332 * @ctx: MMU context associated with owning VM context.
2333 * @sgt: Scatter gather table containing pages pinned for use by this context.
2334 * @sgt_offset: Start offset of the requested device-virtual memory mapping.
2335 * @size: Size in bytes of the requested device-virtual memory mapping. For an
2336 * unmapping, this should be zero so that no page tables are allocated.
2337 *
2338 * Returns:
2339 *  * Newly created MMU op context object on success, or
2340 *  * -%ENOMEM if no memory is available,
2341 *  * Any error code returned by pvr_page_table_l2_init().
2342 */
2343struct pvr_mmu_op_context *
2344pvr_mmu_op_context_create(struct pvr_mmu_context *ctx, struct sg_table *sgt,
2345			  u64 sgt_offset, u64 size)
2346{
2347	int err;
2348
2349	struct pvr_mmu_op_context *op_ctx =
2350		kzalloc(sizeof(*op_ctx), GFP_KERNEL);
2351
2352	if (!op_ctx)
2353		return ERR_PTR(-ENOMEM);
2354
2355	op_ctx->mmu_ctx = ctx;
2356	op_ctx->map.sgt = sgt;
2357	op_ctx->map.sgt_offset = sgt_offset;
2358	op_ctx->sync_level_required = PVR_MMU_SYNC_LEVEL_NONE;
2359
2360	if (size) {
2361		/*
2362		 * The number of page table objects we need to prealloc is
2363		 * indicated by the mapping size, start offset and the sizes
2364		 * of the areas mapped per PT or PD. The range calculation is
2365		 * identical to that for the index into a table for a device
2366		 * address, so we reuse those functions here.
2367		 */
2368		const u32 l1_start_idx = pvr_page_table_l2_idx(sgt_offset);
2369		const u32 l1_end_idx = pvr_page_table_l2_idx(sgt_offset + size);
2370		const u32 l1_count = l1_end_idx - l1_start_idx + 1;
2371		const u32 l0_start_idx = pvr_page_table_l1_idx(sgt_offset);
2372		const u32 l0_end_idx = pvr_page_table_l1_idx(sgt_offset + size);
2373		const u32 l0_count = l0_end_idx - l0_start_idx + 1;
2374
2375		/*
2376		 * Alloc and push page table entries until we have enough of
2377		 * each type, ending with linked lists of l0 and l1 entries in
2378		 * reverse order.
2379		 */
2380		for (int i = 0; i < l1_count; i++) {
2381			struct pvr_page_table_l1 *l1_tmp =
2382				pvr_page_table_l1_alloc(ctx);
2383
2384			err = PTR_ERR_OR_ZERO(l1_tmp);
2385			if (err)
2386				goto err_cleanup;
2387
2388			l1_tmp->next_free = op_ctx->map.l1_prealloc_tables;
2389			op_ctx->map.l1_prealloc_tables = l1_tmp;
2390		}
2391
2392		for (int i = 0; i < l0_count; i++) {
2393			struct pvr_page_table_l0 *l0_tmp =
2394				pvr_page_table_l0_alloc(ctx);
2395
2396			err = PTR_ERR_OR_ZERO(l0_tmp);
2397			if (err)
2398				goto err_cleanup;
2399
2400			l0_tmp->next_free = op_ctx->map.l0_prealloc_tables;
2401			op_ctx->map.l0_prealloc_tables = l0_tmp;
2402		}
2403	}
2404
2405	return op_ctx;
2406
2407err_cleanup:
2408	pvr_mmu_op_context_destroy(op_ctx);
2409
2410	return ERR_PTR(err);
2411}
2412
2413/**
2414 * pvr_mmu_op_context_unmap_curr_page() - Unmap pages from a memory context
2415 * starting from the current page of an MMU op context.
2416 * @op_ctx: Target MMU op context pointing at the first page to unmap.
2417 * @nr_pages: Number of pages to unmap.
2418 *
2419 * Return:
2420 *  * 0 on success, or
2421 *  * Any error encountered while advancing @op_ctx.curr_page with
2422 *    pvr_mmu_op_context_next_page() (except -%ENXIO).
2423 */
2424static int
2425pvr_mmu_op_context_unmap_curr_page(struct pvr_mmu_op_context *op_ctx,
2426				   u64 nr_pages)
2427{
2428	int err;
2429
2430	if (nr_pages == 0)
2431		return 0;
2432
2433	/*
2434	 * Destroy first page outside loop, as it doesn't require a page
2435	 * advance beforehand. If the L0 page table reference in
2436	 * @op_ctx.curr_page is %NULL, there cannot be a mapped page at
2437	 * @op_ctx.curr_page (so skip ahead).
2438	 */
2439	if (op_ctx->curr_page.l0_table)
2440		pvr_page_destroy(op_ctx);
2441
2442	for (u64 page = 1; page < nr_pages; ++page) {
2443		err = pvr_mmu_op_context_next_page(op_ctx, false);
2444		/*
2445		 * If the page table tree structure at @op_ctx.curr_page is
2446		 * incomplete, skip ahead. We don't care about unmapping pages
2447		 * that cannot exist.
2448		 *
2449		 * FIXME: This could be made more efficient by jumping ahead
2450		 * using pvr_mmu_op_context_set_curr_page().
2451		 */
2452		if (err == -ENXIO)
2453			continue;
2454		else if (err)
2455			return err;
2456
2457		pvr_page_destroy(op_ctx);
2458	}
2459
2460	return 0;
2461}
2462
2463/**
2464 * pvr_mmu_unmap() - Unmap pages from a memory context.
2465 * @op_ctx: Target MMU op context.
2466 * @device_addr: First device-virtual address to unmap.
2467 * @size: Size in bytes to unmap.
2468 *
2469 * The total amount of device-virtual memory unmapped is
2470 * @nr_pages * %PVR_DEVICE_PAGE_SIZE.
2471 *
2472 * Returns:
2473 *  * 0 on success, or
2474 *  * Any error code returned by pvr_page_table_ptr_init(), or
2475 *  * Any error code returned by pvr_page_table_ptr_unmap().
2476 */
2477int pvr_mmu_unmap(struct pvr_mmu_op_context *op_ctx, u64 device_addr, u64 size)
2478{
2479	int err = pvr_mmu_op_context_set_curr_page(op_ctx, device_addr, false);
2480
2481	if (err)
2482		return err;
2483
2484	return pvr_mmu_op_context_unmap_curr_page(op_ctx,
2485						  size >> PVR_DEVICE_PAGE_SHIFT);
2486}
2487
2488/**
2489 * pvr_mmu_map_sgl() - Map part of a scatter-gather table entry to
2490 * device-virtual memory.
2491 * @op_ctx: Target MMU op context pointing to the first page that should be
2492 * mapped.
2493 * @sgl: Target scatter-gather table entry.
2494 * @offset: Offset into @sgl to map from. Must result in a starting address
2495 * from @sgl which is CPU page-aligned.
2496 * @size: Size of the memory to be mapped in bytes. Must be a non-zero multiple
2497 * of the device page size.
2498 * @page_flags: Page options to be applied to every device-virtual memory page
2499 * in the created mapping.
2500 *
2501 * Return:
2502 *  * 0 on success,
2503 *  * -%EINVAL if the range specified by @offset and @size is not completely
2504 *    within @sgl, or
2505 *  * Any error encountered while creating a page with pvr_page_create(), or
2506 *  * Any error encountered while advancing @op_ctx.curr_page with
2507 *    pvr_mmu_op_context_next_page().
2508 */
2509static int
2510pvr_mmu_map_sgl(struct pvr_mmu_op_context *op_ctx, struct scatterlist *sgl,
2511		u64 offset, u64 size, struct pvr_page_flags_raw page_flags)
2512{
2513	const unsigned int pages = size >> PVR_DEVICE_PAGE_SHIFT;
2514	dma_addr_t dma_addr = sg_dma_address(sgl) + offset;
2515	const unsigned int dma_len = sg_dma_len(sgl);
2516	struct pvr_page_table_ptr ptr_copy;
2517	unsigned int page;
2518	int err;
2519
2520	if (size > dma_len || offset > dma_len - size)
2521		return -EINVAL;
2522
2523	/*
2524	 * Before progressing, save a copy of the start pointer so we can use
2525	 * it again if we enter an error state and have to destroy pages.
2526	 */
2527	memcpy(&ptr_copy, &op_ctx->curr_page, sizeof(ptr_copy));
2528
2529	/*
2530	 * Create first page outside loop, as it doesn't require a page advance
2531	 * beforehand.
2532	 */
2533	err = pvr_page_create(op_ctx, dma_addr, page_flags);
2534	if (err)
2535		return err;
2536
2537	for (page = 1; page < pages; ++page) {
2538		err = pvr_mmu_op_context_next_page(op_ctx, true);
2539		if (err)
2540			goto err_destroy_pages;
2541
2542		dma_addr += PVR_DEVICE_PAGE_SIZE;
2543
2544		err = pvr_page_create(op_ctx, dma_addr, page_flags);
2545		if (err)
2546			goto err_destroy_pages;
2547	}
2548
2549	return 0;
2550
2551err_destroy_pages:
2552	memcpy(&op_ctx->curr_page, &ptr_copy, sizeof(op_ctx->curr_page));
2553	err = pvr_mmu_op_context_unmap_curr_page(op_ctx, page);
2554
2555	return err;
2556}
2557
2558/**
2559 * pvr_mmu_map() - Map an object's virtual memory to physical memory.
2560 * @op_ctx: Target MMU op context.
2561 * @size: Size of memory to be mapped in bytes. Must be a non-zero multiple
2562 * of the device page size.
2563 * @flags: Flags from pvr_gem_object associated with the mapping.
2564 * @device_addr: Virtual device address to map to. Must be device page-aligned.
2565 *
2566 * Returns:
2567 *  * 0 on success, or
2568 *  * Any error code returned by pvr_page_table_ptr_init(), or
2569 *  * Any error code returned by pvr_mmu_map_sgl(), or
2570 *  * Any error code returned by pvr_page_table_ptr_next_page().
2571 */
2572int pvr_mmu_map(struct pvr_mmu_op_context *op_ctx, u64 size, u64 flags,
2573		u64 device_addr)
2574{
2575	struct pvr_page_table_ptr ptr_copy;
2576	struct pvr_page_flags_raw flags_raw;
2577	struct scatterlist *sgl;
2578	u64 mapped_size = 0;
2579	unsigned int count;
2580	int err;
2581
2582	if (!size)
2583		return 0;
2584
2585	if ((op_ctx->map.sgt_offset | size) & ~PVR_DEVICE_PAGE_MASK)
2586		return -EINVAL;
2587
2588	err = pvr_mmu_op_context_set_curr_page(op_ctx, device_addr, true);
2589	if (err)
2590		return -EINVAL;
2591
2592	memcpy(&ptr_copy, &op_ctx->curr_page, sizeof(ptr_copy));
2593
2594	flags_raw = pvr_page_flags_raw_create(false, false,
2595					      flags & DRM_PVR_BO_BYPASS_DEVICE_CACHE,
2596					      flags & DRM_PVR_BO_PM_FW_PROTECT);
2597
2598	/* Map scatter gather table */
2599	for_each_sgtable_dma_sg(op_ctx->map.sgt, sgl, count) {
2600		const size_t sgl_len = sg_dma_len(sgl);
2601		u64 sgl_offset, map_sgl_len;
2602
2603		if (sgl_len <= op_ctx->map.sgt_offset) {
2604			op_ctx->map.sgt_offset -= sgl_len;
2605			continue;
2606		}
2607
2608		sgl_offset = op_ctx->map.sgt_offset;
2609		map_sgl_len = min_t(u64, sgl_len - sgl_offset, size - mapped_size);
2610
2611		err = pvr_mmu_map_sgl(op_ctx, sgl, sgl_offset, map_sgl_len,
2612				      flags_raw);
2613		if (err)
2614			break;
2615
2616		/*
2617		 * Flag the L0 page table as requiring a flush when the MMU op
2618		 * context is destroyed.
2619		 */
2620		pvr_mmu_op_context_require_sync(op_ctx, PVR_MMU_SYNC_LEVEL_0);
2621
2622		op_ctx->map.sgt_offset = 0;
2623		mapped_size += map_sgl_len;
2624
2625		if (mapped_size >= size)
2626			break;
2627
2628		err = pvr_mmu_op_context_next_page(op_ctx, true);
2629		if (err)
2630			break;
2631	}
2632
2633	if (err && mapped_size) {
2634		memcpy(&op_ctx->curr_page, &ptr_copy, sizeof(op_ctx->curr_page));
2635		pvr_mmu_op_context_unmap_curr_page(op_ctx,
2636						   mapped_size >> PVR_DEVICE_PAGE_SHIFT);
2637	}
2638
2639	return err;
2640}