1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2022, Oracle and/or its affiliates.
4 * Copyright (c) 2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved
5 */
6 #include <linux/iova_bitmap.h>
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/highmem.h>
10
11 #define BITS_PER_PAGE (PAGE_SIZE * BITS_PER_BYTE)
12
13 /*
14 * struct iova_bitmap_map - A bitmap representing an IOVA range
15 *
16 * Main data structure for tracking mapped user pages of bitmap data.
17 *
18 * For example, for something recording dirty IOVAs, it will be provided a
19 * struct iova_bitmap structure, as a general structure for iterating the
20 * total IOVA range. The struct iova_bitmap_map, though, represents the
21 * subset of said IOVA space that is pinned by its parent structure (struct
22 * iova_bitmap).
23 *
24 * The user does not need to exact location of the bits in the bitmap.
25 * From user perspective the only API available is iova_bitmap_set() which
26 * records the IOVA *range* in the bitmap by setting the corresponding
27 * bits.
28 *
29 * The bitmap is an array of u64 whereas each bit represents an IOVA of
30 * range of (1 << pgshift). Thus formula for the bitmap data to be set is:
31 *
32 * data[(iova / page_size) / 64] & (1ULL << (iova % 64))
33 */
34 struct iova_bitmap_map {
35 /* base IOVA representing bit 0 of the first page */
36 unsigned long iova;
37
38 /* page size order that each bit granules to */
39 unsigned long pgshift;
40
41 /* page offset of the first user page pinned */
42 unsigned long pgoff;
43
44 /* number of pages pinned */
45 unsigned long npages;
46
47 /* pinned pages representing the bitmap data */
48 struct page **pages;
49 };
50
51 /*
52 * struct iova_bitmap - The IOVA bitmap object
53 *
54 * Main data structure for iterating over the bitmap data.
55 *
56 * Abstracts the pinning work and iterates in IOVA ranges.
57 * It uses a windowing scheme and pins the bitmap in relatively
58 * big ranges e.g.
59 *
60 * The bitmap object uses one base page to store all the pinned pages
61 * pointers related to the bitmap. For sizeof(struct page*) == 8 it stores
62 * 512 struct page pointers which, if the base page size is 4K, it means
63 * 2M of bitmap data is pinned at a time. If the iova_bitmap page size is
64 * also 4K then the range window to iterate is 64G.
65 *
66 * For example iterating on a total IOVA range of 4G..128G, it will walk
67 * through this set of ranges:
68 *
69 * 4G - 68G-1 (64G)
70 * 68G - 128G-1 (64G)
71 *
72 * An example of the APIs on how to use/iterate over the IOVA bitmap:
73 *
74 * bitmap = iova_bitmap_alloc(iova, length, page_size, data);
75 * if (IS_ERR(bitmap))
76 * return PTR_ERR(bitmap);
77 *
78 * ret = iova_bitmap_for_each(bitmap, arg, dirty_reporter_fn);
79 *
80 * iova_bitmap_free(bitmap);
81 *
82 * Each iteration of the @dirty_reporter_fn is called with a unique @iova
83 * and @length argument, indicating the current range available through the
84 * iova_bitmap. The @dirty_reporter_fn uses iova_bitmap_set() to mark dirty
85 * areas (@iova_length) within that provided range, as following:
86 *
87 * iova_bitmap_set(bitmap, iova, iova_length);
88 *
89 * The internals of the object uses an index @mapped_base_index that indexes
90 * which u64 word of the bitmap is mapped, up to @mapped_total_index.
91 * Those keep being incremented until @mapped_total_index is reached while
92 * mapping up to PAGE_SIZE / sizeof(struct page*) maximum of pages.
93 *
94 * The IOVA bitmap is usually located on what tracks DMA mapped ranges or
95 * some form of IOVA range tracking that co-relates to the user passed
96 * bitmap.
97 */
98 struct iova_bitmap {
99 /* IOVA range representing the currently mapped bitmap data */
100 struct iova_bitmap_map mapped;
101
102 /* userspace address of the bitmap */
103 u8 __user *bitmap;
104
105 /* u64 index that @mapped points to */
106 unsigned long mapped_base_index;
107
108 /* how many u64 can we walk in total */
109 unsigned long mapped_total_index;
110
111 /* base IOVA of the whole bitmap */
112 unsigned long iova;
113
114 /* length of the IOVA range for the whole bitmap */
115 size_t length;
116 };
117
118 /*
119 * Converts a relative IOVA to a bitmap index.
120 * This function provides the index into the u64 array (bitmap::bitmap)
121 * for a given IOVA offset.
122 * Relative IOVA means relative to the bitmap::mapped base IOVA
123 * (stored in mapped::iova). All computations in this file are done using
124 * relative IOVAs and thus avoid an extra subtraction against mapped::iova.
125 * The user API iova_bitmap_set() always uses a regular absolute IOVAs.
126 */
iova_bitmap_offset_to_index(struct iova_bitmap * bitmap,unsigned long iova)127 static unsigned long iova_bitmap_offset_to_index(struct iova_bitmap *bitmap,
128 unsigned long iova)
129 {
130 unsigned long pgsize = 1 << bitmap->mapped.pgshift;
131
132 return iova / (BITS_PER_TYPE(*bitmap->bitmap) * pgsize);
133 }
134
135 /*
136 * Converts a bitmap index to a *relative* IOVA.
137 */
iova_bitmap_index_to_offset(struct iova_bitmap * bitmap,unsigned long index)138 static unsigned long iova_bitmap_index_to_offset(struct iova_bitmap *bitmap,
139 unsigned long index)
140 {
141 unsigned long pgshift = bitmap->mapped.pgshift;
142
143 return (index * BITS_PER_TYPE(*bitmap->bitmap)) << pgshift;
144 }
145
146 /*
147 * Returns the base IOVA of the mapped range.
148 */
iova_bitmap_mapped_iova(struct iova_bitmap * bitmap)149 static unsigned long iova_bitmap_mapped_iova(struct iova_bitmap *bitmap)
150 {
151 unsigned long skip = bitmap->mapped_base_index;
152
153 return bitmap->iova + iova_bitmap_index_to_offset(bitmap, skip);
154 }
155
156 /*
157 * Pins the bitmap user pages for the current range window.
158 * This is internal to IOVA bitmap and called when advancing the
159 * index (@mapped_base_index) or allocating the bitmap.
160 */
iova_bitmap_get(struct iova_bitmap * bitmap)161 static int iova_bitmap_get(struct iova_bitmap *bitmap)
162 {
163 struct iova_bitmap_map *mapped = &bitmap->mapped;
164 unsigned long npages;
165 u8 __user *addr;
166 long ret;
167
168 /*
169 * @mapped_base_index is the index of the currently mapped u64 words
170 * that we have access. Anything before @mapped_base_index is not
171 * mapped. The range @mapped_base_index .. @mapped_total_index-1 is
172 * mapped but capped at a maximum number of pages.
173 */
174 npages = DIV_ROUND_UP((bitmap->mapped_total_index -
175 bitmap->mapped_base_index) *
176 sizeof(*bitmap->bitmap), PAGE_SIZE);
177
178 /*
179 * Bitmap address to be pinned is calculated via pointer arithmetic
180 * with bitmap u64 word index.
181 */
182 addr = bitmap->bitmap + bitmap->mapped_base_index;
183
184 /*
185 * We always cap at max number of 'struct page' a base page can fit.
186 * This is, for example, on x86 means 2M of bitmap data max.
187 */
188 npages = min(npages + !!offset_in_page(addr),
189 PAGE_SIZE / sizeof(struct page *));
190
191 ret = pin_user_pages_fast((unsigned long)addr, npages,
192 FOLL_WRITE, mapped->pages);
193 if (ret <= 0)
194 return -EFAULT;
195
196 mapped->npages = (unsigned long)ret;
197 /* Base IOVA where @pages point to i.e. bit 0 of the first page */
198 mapped->iova = iova_bitmap_mapped_iova(bitmap);
199
200 /*
201 * offset of the page where pinned pages bit 0 is located.
202 * This handles the case where the bitmap is not PAGE_SIZE
203 * aligned.
204 */
205 mapped->pgoff = offset_in_page(addr);
206 return 0;
207 }
208
209 /*
210 * Unpins the bitmap user pages and clears @npages
211 * (un)pinning is abstracted from API user and it's done when advancing
212 * the index or freeing the bitmap.
213 */
iova_bitmap_put(struct iova_bitmap * bitmap)214 static void iova_bitmap_put(struct iova_bitmap *bitmap)
215 {
216 struct iova_bitmap_map *mapped = &bitmap->mapped;
217
218 if (mapped->npages) {
219 unpin_user_pages(mapped->pages, mapped->npages);
220 mapped->npages = 0;
221 }
222 }
223
224 /**
225 * iova_bitmap_alloc() - Allocates an IOVA bitmap object
226 * @iova: Start address of the IOVA range
227 * @length: Length of the IOVA range
228 * @page_size: Page size of the IOVA bitmap. It defines what each bit
229 * granularity represents
230 * @data: Userspace address of the bitmap
231 *
232 * Allocates an IOVA object and initializes all its fields including the
233 * first user pages of @data.
234 *
235 * Return: A pointer to a newly allocated struct iova_bitmap
236 * or ERR_PTR() on error.
237 */
iova_bitmap_alloc(unsigned long iova,size_t length,unsigned long page_size,u64 __user * data)238 struct iova_bitmap *iova_bitmap_alloc(unsigned long iova, size_t length,
239 unsigned long page_size, u64 __user *data)
240 {
241 struct iova_bitmap_map *mapped;
242 struct iova_bitmap *bitmap;
243 int rc;
244
245 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
246 if (!bitmap)
247 return ERR_PTR(-ENOMEM);
248
249 mapped = &bitmap->mapped;
250 mapped->pgshift = __ffs(page_size);
251 bitmap->bitmap = (u8 __user *)data;
252 bitmap->mapped_total_index =
253 iova_bitmap_offset_to_index(bitmap, length - 1) + 1;
254 bitmap->iova = iova;
255 bitmap->length = length;
256 mapped->iova = iova;
257 mapped->pages = (struct page **)__get_free_page(GFP_KERNEL);
258 if (!mapped->pages) {
259 rc = -ENOMEM;
260 goto err;
261 }
262
263 rc = iova_bitmap_get(bitmap);
264 if (rc)
265 goto err;
266 return bitmap;
267
268 err:
269 iova_bitmap_free(bitmap);
270 return ERR_PTR(rc);
271 }
272
273 /**
274 * iova_bitmap_free() - Frees an IOVA bitmap object
275 * @bitmap: IOVA bitmap to free
276 *
277 * It unpins and releases pages array memory and clears any leftover
278 * state.
279 */
iova_bitmap_free(struct iova_bitmap * bitmap)280 void iova_bitmap_free(struct iova_bitmap *bitmap)
281 {
282 struct iova_bitmap_map *mapped = &bitmap->mapped;
283
284 iova_bitmap_put(bitmap);
285
286 if (mapped->pages) {
287 free_page((unsigned long)mapped->pages);
288 mapped->pages = NULL;
289 }
290
291 kfree(bitmap);
292 }
293
294 /*
295 * Returns the remaining bitmap indexes from mapped_total_index to process for
296 * the currently pinned bitmap pages.
297 */
iova_bitmap_mapped_remaining(struct iova_bitmap * bitmap)298 static unsigned long iova_bitmap_mapped_remaining(struct iova_bitmap *bitmap)
299 {
300 unsigned long remaining, bytes;
301
302 bytes = (bitmap->mapped.npages << PAGE_SHIFT) - bitmap->mapped.pgoff;
303
304 remaining = bitmap->mapped_total_index - bitmap->mapped_base_index;
305 remaining = min_t(unsigned long, remaining,
306 DIV_ROUND_UP(bytes, sizeof(*bitmap->bitmap)));
307
308 return remaining;
309 }
310
311 /*
312 * Returns the length of the mapped IOVA range.
313 */
iova_bitmap_mapped_length(struct iova_bitmap * bitmap)314 static unsigned long iova_bitmap_mapped_length(struct iova_bitmap *bitmap)
315 {
316 unsigned long max_iova = bitmap->iova + bitmap->length - 1;
317 unsigned long iova = iova_bitmap_mapped_iova(bitmap);
318 unsigned long remaining;
319
320 /*
321 * iova_bitmap_mapped_remaining() returns a number of indexes which
322 * when converted to IOVA gives us a max length that the bitmap
323 * pinned data can cover. Afterwards, that is capped to
324 * only cover the IOVA range in @bitmap::iova .. @bitmap::length.
325 */
326 remaining = iova_bitmap_index_to_offset(bitmap,
327 iova_bitmap_mapped_remaining(bitmap));
328
329 if (iova + remaining - 1 > max_iova)
330 remaining -= ((iova + remaining - 1) - max_iova);
331
332 return remaining;
333 }
334
335 /*
336 * Returns true if there's not more data to iterate.
337 */
iova_bitmap_done(struct iova_bitmap * bitmap)338 static bool iova_bitmap_done(struct iova_bitmap *bitmap)
339 {
340 return bitmap->mapped_base_index >= bitmap->mapped_total_index;
341 }
342
343 /*
344 * Advances to the next range, releases the current pinned
345 * pages and pins the next set of bitmap pages.
346 * Returns 0 on success or otherwise errno.
347 */
iova_bitmap_advance(struct iova_bitmap * bitmap)348 static int iova_bitmap_advance(struct iova_bitmap *bitmap)
349 {
350 unsigned long iova = iova_bitmap_mapped_length(bitmap) - 1;
351 unsigned long count = iova_bitmap_offset_to_index(bitmap, iova) + 1;
352
353 bitmap->mapped_base_index += count;
354
355 iova_bitmap_put(bitmap);
356 if (iova_bitmap_done(bitmap))
357 return 0;
358
359 /* When advancing the index we pin the next set of bitmap pages */
360 return iova_bitmap_get(bitmap);
361 }
362
363 /**
364 * iova_bitmap_for_each() - Iterates over the bitmap
365 * @bitmap: IOVA bitmap to iterate
366 * @opaque: Additional argument to pass to the callback
367 * @fn: Function that gets called for each IOVA range
368 *
369 * Helper function to iterate over bitmap data representing a portion of IOVA
370 * space. It hides the complexity of iterating bitmaps and translating the
371 * mapped bitmap user pages into IOVA ranges to process.
372 *
373 * Return: 0 on success, and an error on failure either upon
374 * iteration or when the callback returns an error.
375 */
iova_bitmap_for_each(struct iova_bitmap * bitmap,void * opaque,iova_bitmap_fn_t fn)376 int iova_bitmap_for_each(struct iova_bitmap *bitmap, void *opaque,
377 iova_bitmap_fn_t fn)
378 {
379 int ret = 0;
380
381 for (; !iova_bitmap_done(bitmap) && !ret;
382 ret = iova_bitmap_advance(bitmap)) {
383 ret = fn(bitmap, iova_bitmap_mapped_iova(bitmap),
384 iova_bitmap_mapped_length(bitmap), opaque);
385 if (ret)
386 break;
387 }
388
389 return ret;
390 }
391
392 /**
393 * iova_bitmap_set() - Records an IOVA range in bitmap
394 * @bitmap: IOVA bitmap
395 * @iova: IOVA to start
396 * @length: IOVA range length
397 *
398 * Set the bits corresponding to the range [iova .. iova+length-1] in
399 * the user bitmap.
400 *
401 */
iova_bitmap_set(struct iova_bitmap * bitmap,unsigned long iova,size_t length)402 void iova_bitmap_set(struct iova_bitmap *bitmap,
403 unsigned long iova, size_t length)
404 {
405 struct iova_bitmap_map *mapped = &bitmap->mapped;
406 unsigned long cur_bit = ((iova - mapped->iova) >>
407 mapped->pgshift) + mapped->pgoff * BITS_PER_BYTE;
408 unsigned long last_bit = (((iova + length - 1) - mapped->iova) >>
409 mapped->pgshift) + mapped->pgoff * BITS_PER_BYTE;
410 unsigned long last_page_idx = mapped->npages - 1;
411
412 do {
413 unsigned int page_idx = cur_bit / BITS_PER_PAGE;
414 unsigned int offset = cur_bit % BITS_PER_PAGE;
415 unsigned int nbits = min(BITS_PER_PAGE - offset,
416 last_bit - cur_bit + 1);
417 void *kaddr;
418
419 if (unlikely(page_idx > last_page_idx))
420 break;
421
422 kaddr = kmap_local_page(mapped->pages[page_idx]);
423 bitmap_set(kaddr, offset, nbits);
424 kunmap_local(kaddr);
425 cur_bit += nbits;
426 } while (cur_bit <= last_bit);
427 }
428 EXPORT_SYMBOL_GPL(iova_bitmap_set);
429