Blame view
mm/hmm.c
16.6 KB
c942fddf8
|
1 |
// SPDX-License-Identifier: GPL-2.0-or-later |
133ff0eac
|
2 3 4 |
/* * Copyright 2013 Red Hat Inc. * |
f813f2197
|
5 |
* Authors: Jérôme Glisse <jglisse@redhat.com> |
133ff0eac
|
6 7 8 9 10 |
*/ /* * Refer to include/linux/hmm.h for information about heterogeneous memory * management or HMM for short. */ |
a520110e4
|
11 |
#include <linux/pagewalk.h> |
133ff0eac
|
12 |
#include <linux/hmm.h> |
858b54dab
|
13 |
#include <linux/init.h> |
da4c3c735
|
14 15 |
#include <linux/rmap.h> #include <linux/swap.h> |
133ff0eac
|
16 17 |
#include <linux/slab.h> #include <linux/sched.h> |
4ef589dc9
|
18 19 |
#include <linux/mmzone.h> #include <linux/pagemap.h> |
da4c3c735
|
20 21 |
#include <linux/swapops.h> #include <linux/hugetlb.h> |
4ef589dc9
|
22 |
#include <linux/memremap.h> |
c8a53b2db
|
23 |
#include <linux/sched/mm.h> |
7b2d55d2c
|
24 |
#include <linux/jump_label.h> |
55c0ece82
|
25 |
#include <linux/dma-mapping.h> |
c0b124054
|
26 |
#include <linux/mmu_notifier.h> |
4ef589dc9
|
27 |
#include <linux/memory_hotplug.h> |
74eee180b
|
28 29 30 |
struct hmm_vma_walk { struct hmm_range *range; unsigned long last; |
74eee180b
|
31 |
}; |
a3eb13c15
|
32 33 34 35 36 |
enum { HMM_NEED_FAULT = 1 << 0, HMM_NEED_WRITE_FAULT = 1 << 1, HMM_NEED_ALL_BITS = HMM_NEED_FAULT | HMM_NEED_WRITE_FAULT, }; |
d28c2c9a4
|
37 |
static int hmm_pfns_fill(unsigned long addr, unsigned long end, |
2733ea144
|
38 |
struct hmm_range *range, unsigned long cpu_flags) |
da4c3c735
|
39 |
{ |
2733ea144
|
40 |
unsigned long i = (addr - range->start) >> PAGE_SHIFT; |
da4c3c735
|
41 |
|
da4c3c735
|
42 |
for (; addr < end; addr += PAGE_SIZE, i++) |
2733ea144
|
43 |
range->hmm_pfns[i] = cpu_flags; |
da4c3c735
|
44 45 |
return 0; } |
5504ed296
|
46 |
/* |
f8c888a30
|
47 |
* hmm_vma_fault() - fault in a range lacking valid pmd or pte(s) |
d2e8d5511
|
48 |
* @addr: range virtual start address (inclusive) |
5504ed296
|
49 |
* @end: range virtual end address (exclusive) |
a3eb13c15
|
50 |
* @required_fault: HMM_NEED_* flags |
5504ed296
|
51 |
* @walk: mm_walk structure |
f8c888a30
|
52 |
* Return: -EBUSY after page fault, or page fault error |
5504ed296
|
53 54 55 56 |
* * This function will be called whenever pmd_none() or pte_none() returns true, * or whenever there is no page directory covering the virtual address range. */ |
f8c888a30
|
57 |
static int hmm_vma_fault(unsigned long addr, unsigned long end, |
a3eb13c15
|
58 |
unsigned int required_fault, struct mm_walk *walk) |
da4c3c735
|
59 |
{ |
74eee180b
|
60 |
struct hmm_vma_walk *hmm_vma_walk = walk->private; |
5a0c38d30
|
61 |
struct vm_area_struct *vma = walk->vma; |
5a0c38d30
|
62 |
unsigned int fault_flags = FAULT_FLAG_REMOTE; |
da4c3c735
|
63 |
|
a3eb13c15
|
64 |
WARN_ON_ONCE(!required_fault); |
74eee180b
|
65 |
hmm_vma_walk->last = addr; |
63d5066f6
|
66 |
|
a3eb13c15
|
67 |
if (required_fault & HMM_NEED_WRITE_FAULT) { |
5a0c38d30
|
68 69 70 |
if (!(vma->vm_flags & VM_WRITE)) return -EPERM; fault_flags |= FAULT_FLAG_WRITE; |
74eee180b
|
71 |
} |
53bfe17ff
|
72 |
for (; addr < end; addr += PAGE_SIZE) |
bce617ede
|
73 74 |
if (handle_mm_fault(vma, addr, fault_flags, NULL) & VM_FAULT_ERROR) |
53bfe17ff
|
75 |
return -EFAULT; |
f8c888a30
|
76 |
return -EBUSY; |
2aee09d8c
|
77 |
} |
a3eb13c15
|
78 |
static unsigned int hmm_pte_need_fault(const struct hmm_vma_walk *hmm_vma_walk, |
2733ea144
|
79 80 |
unsigned long pfn_req_flags, unsigned long cpu_flags) |
2aee09d8c
|
81 |
{ |
f88a1e90c
|
82 |
struct hmm_range *range = hmm_vma_walk->range; |
023a019a9
|
83 84 85 |
/* * So we not only consider the individual per page request we also * consider the default flags requested for the range. The API can |
d2e8d5511
|
86 87 88 |
* be used 2 ways. The first one where the HMM user coalesces * multiple page faults into one request and sets flags per pfn for * those faults. The second one where the HMM user wants to pre- |
023a019a9
|
89 90 91 92 |
* fault a range with specific flags. For the latter one it is a * waste to have the user pre-fill the pfn arrays with a default * flags value. */ |
2733ea144
|
93 94 |
pfn_req_flags &= range->pfn_flags_mask; pfn_req_flags |= range->default_flags; |
023a019a9
|
95 |
|
2aee09d8c
|
96 |
/* We aren't ask to do anything ... */ |
2733ea144
|
97 |
if (!(pfn_req_flags & HMM_PFN_REQ_FAULT)) |
a3eb13c15
|
98 |
return 0; |
f88a1e90c
|
99 |
|
f88a1e90c
|
100 |
/* Need to write fault ? */ |
2733ea144
|
101 102 |
if ((pfn_req_flags & HMM_PFN_REQ_WRITE) && !(cpu_flags & HMM_PFN_WRITE)) |
a3eb13c15
|
103 104 105 |
return HMM_NEED_FAULT | HMM_NEED_WRITE_FAULT; /* If CPU page table is not valid then we need to fault */ |
2733ea144
|
106 |
if (!(cpu_flags & HMM_PFN_VALID)) |
a3eb13c15
|
107 108 |
return HMM_NEED_FAULT; return 0; |
2aee09d8c
|
109 |
} |
a3eb13c15
|
110 111 |
static unsigned int hmm_range_need_fault(const struct hmm_vma_walk *hmm_vma_walk, |
2733ea144
|
112 113 |
const unsigned long hmm_pfns[], unsigned long npages, unsigned long cpu_flags) |
2aee09d8c
|
114 |
{ |
6bfef2f91
|
115 |
struct hmm_range *range = hmm_vma_walk->range; |
a3eb13c15
|
116 |
unsigned int required_fault = 0; |
2aee09d8c
|
117 |
unsigned long i; |
6bfef2f91
|
118 119 120 121 122 123 |
/* * If the default flags do not request to fault pages, and the mask does * not allow for individual pages to be faulted, then * hmm_pte_need_fault() will always return 0. */ if (!((range->default_flags | range->pfn_flags_mask) & |
2733ea144
|
124 |
HMM_PFN_REQ_FAULT)) |
a3eb13c15
|
125 |
return 0; |
2aee09d8c
|
126 127 |
for (i = 0; i < npages; ++i) { |
2733ea144
|
128 129 |
required_fault |= hmm_pte_need_fault(hmm_vma_walk, hmm_pfns[i], cpu_flags); |
a3eb13c15
|
130 131 |
if (required_fault == HMM_NEED_ALL_BITS) return required_fault; |
2aee09d8c
|
132 |
} |
a3eb13c15
|
133 |
return required_fault; |
2aee09d8c
|
134 135 136 |
} static int hmm_vma_walk_hole(unsigned long addr, unsigned long end, |
b7a16c7ad
|
137 |
__always_unused int depth, struct mm_walk *walk) |
2aee09d8c
|
138 139 140 |
{ struct hmm_vma_walk *hmm_vma_walk = walk->private; struct hmm_range *range = hmm_vma_walk->range; |
a3eb13c15
|
141 |
unsigned int required_fault; |
2aee09d8c
|
142 |
unsigned long i, npages; |
2733ea144
|
143 |
unsigned long *hmm_pfns; |
2aee09d8c
|
144 145 146 |
i = (addr - range->start) >> PAGE_SHIFT; npages = (end - addr) >> PAGE_SHIFT; |
2733ea144
|
147 148 149 |
hmm_pfns = &range->hmm_pfns[i]; required_fault = hmm_range_need_fault(hmm_vma_walk, hmm_pfns, npages, 0); |
bd5d3587b
|
150 151 152 153 154 |
if (!walk->vma) { if (required_fault) return -EFAULT; return hmm_pfns_fill(addr, end, range, HMM_PFN_ERROR); } |
a3eb13c15
|
155 156 |
if (required_fault) return hmm_vma_fault(addr, end, required_fault, walk); |
2733ea144
|
157 |
return hmm_pfns_fill(addr, end, range, 0); |
2aee09d8c
|
158 |
} |
3b50a6e53
|
159 160 161 162 |
static inline unsigned long hmm_pfn_flags_order(unsigned long order) { return order << HMM_PFN_ORDER_SHIFT; } |
2733ea144
|
163 164 |
static inline unsigned long pmd_to_hmm_pfn_flags(struct hmm_range *range, pmd_t pmd) |
2aee09d8c
|
165 166 167 |
{ if (pmd_protnone(pmd)) return 0; |
3b50a6e53
|
168 169 170 |
return (pmd_write(pmd) ? (HMM_PFN_VALID | HMM_PFN_WRITE) : HMM_PFN_VALID) | hmm_pfn_flags_order(PMD_SHIFT - PAGE_SHIFT); |
da4c3c735
|
171 |
} |
992de9a8b
|
172 |
#ifdef CONFIG_TRANSPARENT_HUGEPAGE |
9d3973d60
|
173 |
static int hmm_vma_handle_pmd(struct mm_walk *walk, unsigned long addr, |
2733ea144
|
174 175 |
unsigned long end, unsigned long hmm_pfns[], pmd_t pmd) |
9d3973d60
|
176 |
{ |
53f5c3f48
|
177 |
struct hmm_vma_walk *hmm_vma_walk = walk->private; |
f88a1e90c
|
178 |
struct hmm_range *range = hmm_vma_walk->range; |
2aee09d8c
|
179 |
unsigned long pfn, npages, i; |
a3eb13c15
|
180 |
unsigned int required_fault; |
2733ea144
|
181 |
unsigned long cpu_flags; |
53f5c3f48
|
182 |
|
2aee09d8c
|
183 |
npages = (end - addr) >> PAGE_SHIFT; |
f88a1e90c
|
184 |
cpu_flags = pmd_to_hmm_pfn_flags(range, pmd); |
a3eb13c15
|
185 |
required_fault = |
2733ea144
|
186 |
hmm_range_need_fault(hmm_vma_walk, hmm_pfns, npages, cpu_flags); |
a3eb13c15
|
187 188 |
if (required_fault) return hmm_vma_fault(addr, end, required_fault, walk); |
53f5c3f48
|
189 |
|
309f9a4f5
|
190 |
pfn = pmd_pfn(pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); |
068354ade
|
191 |
for (i = 0; addr < end; addr += PAGE_SIZE, i++, pfn++) |
2733ea144
|
192 |
hmm_pfns[i] = pfn | cpu_flags; |
53f5c3f48
|
193 194 |
return 0; } |
9d3973d60
|
195 196 197 |
#else /* CONFIG_TRANSPARENT_HUGEPAGE */ /* stub to allow the code below to compile */ int hmm_vma_handle_pmd(struct mm_walk *walk, unsigned long addr, |
2733ea144
|
198 |
unsigned long end, unsigned long hmm_pfns[], pmd_t pmd); |
9d3973d60
|
199 |
#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
53f5c3f48
|
200 |
|
08ddddda6
|
201 202 203 204 205 206 207 |
static inline bool hmm_is_device_private_entry(struct hmm_range *range, swp_entry_t entry) { return is_device_private_entry(entry) && device_private_entry_to_page(entry)->pgmap->owner == range->dev_private_owner; } |
2733ea144
|
208 209 |
static inline unsigned long pte_to_hmm_pfn_flags(struct hmm_range *range, pte_t pte) |
2aee09d8c
|
210 |
{ |
789c2af88
|
211 |
if (pte_none(pte) || !pte_present(pte) || pte_protnone(pte)) |
2aee09d8c
|
212 |
return 0; |
2733ea144
|
213 |
return pte_write(pte) ? (HMM_PFN_VALID | HMM_PFN_WRITE) : HMM_PFN_VALID; |
2aee09d8c
|
214 |
} |
53f5c3f48
|
215 216 |
static int hmm_vma_handle_pte(struct mm_walk *walk, unsigned long addr, unsigned long end, pmd_t *pmdp, pte_t *ptep, |
2733ea144
|
217 |
unsigned long *hmm_pfn) |
53f5c3f48
|
218 219 |
{ struct hmm_vma_walk *hmm_vma_walk = walk->private; |
f88a1e90c
|
220 |
struct hmm_range *range = hmm_vma_walk->range; |
a3eb13c15
|
221 |
unsigned int required_fault; |
2733ea144
|
222 |
unsigned long cpu_flags; |
53f5c3f48
|
223 |
pte_t pte = *ptep; |
2733ea144
|
224 |
uint64_t pfn_req_flags = *hmm_pfn; |
53f5c3f48
|
225 |
|
53f5c3f48
|
226 |
if (pte_none(pte)) { |
2733ea144
|
227 228 |
required_fault = hmm_pte_need_fault(hmm_vma_walk, pfn_req_flags, 0); |
a3eb13c15
|
229 |
if (required_fault) |
53f5c3f48
|
230 |
goto fault; |
2733ea144
|
231 |
*hmm_pfn = 0; |
53f5c3f48
|
232 233 234 235 236 |
return 0; } if (!pte_present(pte)) { swp_entry_t entry = pte_to_swp_entry(pte); |
53f5c3f48
|
237 |
/* |
0cb80a2fb
|
238 |
* Never fault in device private pages, but just report |
17ffdc482
|
239 |
* the PFN even if not present. |
53f5c3f48
|
240 |
*/ |
08ddddda6
|
241 |
if (hmm_is_device_private_entry(range, entry)) { |
2733ea144
|
242 |
cpu_flags = HMM_PFN_VALID; |
17ffdc482
|
243 |
if (is_write_device_private_entry(entry)) |
2733ea144
|
244 245 246 |
cpu_flags |= HMM_PFN_WRITE; *hmm_pfn = device_private_entry_to_pfn(entry) | cpu_flags; |
53f5c3f48
|
247 248 |
return 0; } |
2733ea144
|
249 250 |
required_fault = hmm_pte_need_fault(hmm_vma_walk, pfn_req_flags, 0); |
846babe85
|
251 |
if (!required_fault) { |
2733ea144
|
252 |
*hmm_pfn = 0; |
53f5c3f48
|
253 |
return 0; |
846babe85
|
254 |
} |
76612d6ce
|
255 256 257 258 259 260 261 262 263 |
if (!non_swap_entry(entry)) goto fault; if (is_migration_entry(entry)) { pte_unmap(ptep); hmm_vma_walk->last = addr; migration_entry_wait(walk->mm, pmdp, addr); return -EBUSY; |
53f5c3f48
|
264 265 266 |
} /* Report error for everything else */ |
dfdc22078
|
267 |
pte_unmap(ptep); |
53f5c3f48
|
268 269 |
return -EFAULT; } |
76612d6ce
|
270 |
cpu_flags = pte_to_hmm_pfn_flags(range, pte); |
2733ea144
|
271 272 |
required_fault = hmm_pte_need_fault(hmm_vma_walk, pfn_req_flags, cpu_flags); |
a3eb13c15
|
273 |
if (required_fault) |
53f5c3f48
|
274 |
goto fault; |
405506274
|
275 276 277 278 279 |
/* * Since each architecture defines a struct page for the zero page, just * fall through and treat it like a normal page. */ if (pte_special(pte) && !is_zero_pfn(pte_pfn(pte))) { |
2733ea144
|
280 |
if (hmm_pte_need_fault(hmm_vma_walk, pfn_req_flags, 0)) { |
dfdc22078
|
281 |
pte_unmap(ptep); |
ac541f250
|
282 283 |
return -EFAULT; } |
2733ea144
|
284 |
*hmm_pfn = HMM_PFN_ERROR; |
405506274
|
285 |
return 0; |
992de9a8b
|
286 |
} |
2733ea144
|
287 |
*hmm_pfn = pte_pfn(pte) | cpu_flags; |
53f5c3f48
|
288 289 290 291 292 |
return 0; fault: pte_unmap(ptep); /* Fault any virtual address we were asked to fault */ |
a3eb13c15
|
293 |
return hmm_vma_fault(addr, end, required_fault, walk); |
53f5c3f48
|
294 |
} |
da4c3c735
|
295 296 297 298 299 |
static int hmm_vma_walk_pmd(pmd_t *pmdp, unsigned long start, unsigned long end, struct mm_walk *walk) { |
74eee180b
|
300 301 |
struct hmm_vma_walk *hmm_vma_walk = walk->private; struct hmm_range *range = hmm_vma_walk->range; |
2733ea144
|
302 303 |
unsigned long *hmm_pfns = &range->hmm_pfns[(start - range->start) >> PAGE_SHIFT]; |
2288a9a68
|
304 305 |
unsigned long npages = (end - start) >> PAGE_SHIFT; unsigned long addr = start; |
da4c3c735
|
306 |
pte_t *ptep; |
d08faca01
|
307 |
pmd_t pmd; |
da4c3c735
|
308 |
|
da4c3c735
|
309 |
again: |
d08faca01
|
310 311 |
pmd = READ_ONCE(*pmdp); if (pmd_none(pmd)) |
b7a16c7ad
|
312 |
return hmm_vma_walk_hole(start, end, -1, walk); |
da4c3c735
|
313 |
|
d08faca01
|
314 |
if (thp_migration_supported() && is_pmd_migration_entry(pmd)) { |
2733ea144
|
315 |
if (hmm_range_need_fault(hmm_vma_walk, hmm_pfns, npages, 0)) { |
d08faca01
|
316 |
hmm_vma_walk->last = addr; |
d2e8d5511
|
317 |
pmd_migration_entry_wait(walk->mm, pmdp); |
73231612d
|
318 |
return -EBUSY; |
d08faca01
|
319 |
} |
2733ea144
|
320 |
return hmm_pfns_fill(start, end, range, 0); |
2288a9a68
|
321 322 323 |
} if (!pmd_present(pmd)) { |
2733ea144
|
324 |
if (hmm_range_need_fault(hmm_vma_walk, hmm_pfns, npages, 0)) |
2288a9a68
|
325 |
return -EFAULT; |
d28c2c9a4
|
326 |
return hmm_pfns_fill(start, end, range, HMM_PFN_ERROR); |
2288a9a68
|
327 |
} |
da4c3c735
|
328 |
|
d08faca01
|
329 |
if (pmd_devmap(pmd) || pmd_trans_huge(pmd)) { |
da4c3c735
|
330 |
/* |
d2e8d5511
|
331 |
* No need to take pmd_lock here, even if some other thread |
da4c3c735
|
332 333 334 |
* is splitting the huge pmd we will get that event through * mmu_notifier callback. * |
d2e8d5511
|
335 |
* So just read pmd value and check again it's a transparent |
da4c3c735
|
336 337 338 339 340 341 342 |
* huge or device mapping one and compute corresponding pfn * values. */ pmd = pmd_read_atomic(pmdp); barrier(); if (!pmd_devmap(pmd) && !pmd_trans_huge(pmd)) goto again; |
74eee180b
|
343 |
|
2733ea144
|
344 |
return hmm_vma_handle_pmd(walk, addr, end, hmm_pfns, pmd); |
da4c3c735
|
345 |
} |
d08faca01
|
346 |
/* |
d2e8d5511
|
347 |
* We have handled all the valid cases above ie either none, migration, |
d08faca01
|
348 349 350 351 |
* huge or transparent huge. At this point either it is a valid pmd * entry pointing to pte directory or it is a bad pmd that will not * recover. */ |
2288a9a68
|
352 |
if (pmd_bad(pmd)) { |
2733ea144
|
353 |
if (hmm_range_need_fault(hmm_vma_walk, hmm_pfns, npages, 0)) |
2288a9a68
|
354 |
return -EFAULT; |
d28c2c9a4
|
355 |
return hmm_pfns_fill(start, end, range, HMM_PFN_ERROR); |
2288a9a68
|
356 |
} |
da4c3c735
|
357 358 |
ptep = pte_offset_map(pmdp, addr); |
2733ea144
|
359 |
for (; addr < end; addr += PAGE_SIZE, ptep++, hmm_pfns++) { |
53f5c3f48
|
360 |
int r; |
74eee180b
|
361 |
|
2733ea144
|
362 |
r = hmm_vma_handle_pte(walk, addr, end, pmdp, ptep, hmm_pfns); |
53f5c3f48
|
363 |
if (r) { |
dfdc22078
|
364 |
/* hmm_vma_handle_pte() did pte_unmap() */ |
53f5c3f48
|
365 |
return r; |
74eee180b
|
366 |
} |
da4c3c735
|
367 368 |
} pte_unmap(ptep - 1); |
da4c3c735
|
369 370 |
return 0; } |
f0b3c45c8
|
371 372 |
#if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && \ defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) |
2733ea144
|
373 374 |
static inline unsigned long pud_to_hmm_pfn_flags(struct hmm_range *range, pud_t pud) |
f0b3c45c8
|
375 376 377 |
{ if (!pud_present(pud)) return 0; |
3b50a6e53
|
378 379 380 |
return (pud_write(pud) ? (HMM_PFN_VALID | HMM_PFN_WRITE) : HMM_PFN_VALID) | hmm_pfn_flags_order(PUD_SHIFT - PAGE_SHIFT); |
f0b3c45c8
|
381 382 383 384 |
} static int hmm_vma_walk_pud(pud_t *pudp, unsigned long start, unsigned long end, struct mm_walk *walk) |
992de9a8b
|
385 386 387 |
{ struct hmm_vma_walk *hmm_vma_walk = walk->private; struct hmm_range *range = hmm_vma_walk->range; |
3afc42363
|
388 |
unsigned long addr = start; |
992de9a8b
|
389 |
pud_t pud; |
3afc42363
|
390 391 392 393 394 395 396 397 |
int ret = 0; spinlock_t *ptl = pud_trans_huge_lock(pudp, walk->vma); if (!ptl) return 0; /* Normally we don't want to split the huge page */ walk->action = ACTION_CONTINUE; |
992de9a8b
|
398 |
|
992de9a8b
|
399 |
pud = READ_ONCE(*pudp); |
3afc42363
|
400 |
if (pud_none(pud)) { |
05fc1df95
|
401 402 |
spin_unlock(ptl); return hmm_vma_walk_hole(start, end, -1, walk); |
3afc42363
|
403 |
} |
992de9a8b
|
404 405 406 |
if (pud_huge(pud) && pud_devmap(pud)) { unsigned long i, npages, pfn; |
a3eb13c15
|
407 |
unsigned int required_fault; |
2733ea144
|
408 409 |
unsigned long *hmm_pfns; unsigned long cpu_flags; |
992de9a8b
|
410 |
|
3afc42363
|
411 |
if (!pud_present(pud)) { |
05fc1df95
|
412 413 |
spin_unlock(ptl); return hmm_vma_walk_hole(start, end, -1, walk); |
3afc42363
|
414 |
} |
992de9a8b
|
415 416 417 |
i = (addr - range->start) >> PAGE_SHIFT; npages = (end - addr) >> PAGE_SHIFT; |
2733ea144
|
418 |
hmm_pfns = &range->hmm_pfns[i]; |
992de9a8b
|
419 420 |
cpu_flags = pud_to_hmm_pfn_flags(range, pud); |
2733ea144
|
421 |
required_fault = hmm_range_need_fault(hmm_vma_walk, hmm_pfns, |
a3eb13c15
|
422 423 |
npages, cpu_flags); if (required_fault) { |
05fc1df95
|
424 |
spin_unlock(ptl); |
a3eb13c15
|
425 |
return hmm_vma_fault(addr, end, required_fault, walk); |
3afc42363
|
426 |
} |
992de9a8b
|
427 |
|
992de9a8b
|
428 |
pfn = pud_pfn(pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); |
068354ade
|
429 |
for (i = 0; i < npages; ++i, ++pfn) |
2733ea144
|
430 |
hmm_pfns[i] = pfn | cpu_flags; |
3afc42363
|
431 |
goto out_unlock; |
992de9a8b
|
432 |
} |
3afc42363
|
433 434 |
/* Ask for the PUD to be split */ walk->action = ACTION_SUBTREE; |
992de9a8b
|
435 |
|
3afc42363
|
436 437 438 |
out_unlock: spin_unlock(ptl); return ret; |
992de9a8b
|
439 |
} |
f0b3c45c8
|
440 441 442 |
#else #define hmm_vma_walk_pud NULL #endif |
992de9a8b
|
443 |
|
251bbe59b
|
444 |
#ifdef CONFIG_HUGETLB_PAGE |
63d5066f6
|
445 446 447 448 |
static int hmm_vma_walk_hugetlb_entry(pte_t *pte, unsigned long hmask, unsigned long start, unsigned long end, struct mm_walk *walk) { |
05c23af4a
|
449 |
unsigned long addr = start, i, pfn; |
63d5066f6
|
450 451 452 |
struct hmm_vma_walk *hmm_vma_walk = walk->private; struct hmm_range *range = hmm_vma_walk->range; struct vm_area_struct *vma = walk->vma; |
a3eb13c15
|
453 |
unsigned int required_fault; |
2733ea144
|
454 455 |
unsigned long pfn_req_flags; unsigned long cpu_flags; |
63d5066f6
|
456 457 |
spinlock_t *ptl; pte_t entry; |
63d5066f6
|
458 |
|
d2e8d5511
|
459 |
ptl = huge_pte_lock(hstate_vma(vma), walk->mm, pte); |
63d5066f6
|
460 |
entry = huge_ptep_get(pte); |
7f08263d9
|
461 |
i = (start - range->start) >> PAGE_SHIFT; |
2733ea144
|
462 |
pfn_req_flags = range->hmm_pfns[i]; |
3b50a6e53
|
463 464 |
cpu_flags = pte_to_hmm_pfn_flags(range, entry) | hmm_pfn_flags_order(huge_page_order(hstate_vma(vma))); |
2733ea144
|
465 466 |
required_fault = hmm_pte_need_fault(hmm_vma_walk, pfn_req_flags, cpu_flags); |
a3eb13c15
|
467 |
if (required_fault) { |
45050692d
|
468 |
spin_unlock(ptl); |
a3eb13c15
|
469 |
return hmm_vma_fault(addr, end, required_fault, walk); |
63d5066f6
|
470 |
} |
05c23af4a
|
471 |
pfn = pte_pfn(entry) + ((start & ~hmask) >> PAGE_SHIFT); |
7f08263d9
|
472 |
for (; addr < end; addr += PAGE_SIZE, i++, pfn++) |
2733ea144
|
473 |
range->hmm_pfns[i] = pfn | cpu_flags; |
63d5066f6
|
474 |
spin_unlock(ptl); |
45050692d
|
475 |
return 0; |
63d5066f6
|
476 |
} |
251bbe59b
|
477 478 479 |
#else #define hmm_vma_walk_hugetlb_entry NULL #endif /* CONFIG_HUGETLB_PAGE */ |
63d5066f6
|
480 |
|
d28c2c9a4
|
481 482 |
static int hmm_vma_walk_test(unsigned long start, unsigned long end, struct mm_walk *walk) |
33cd47dcb
|
483 |
{ |
d28c2c9a4
|
484 485 486 |
struct hmm_vma_walk *hmm_vma_walk = walk->private; struct hmm_range *range = hmm_vma_walk->range; struct vm_area_struct *vma = walk->vma; |
a3eb13c15
|
487 488 489 |
if (!(vma->vm_flags & (VM_IO | VM_PFNMAP | VM_MIXEDMAP)) && vma->vm_flags & VM_READ) return 0; |
d28c2c9a4
|
490 |
/* |
a3eb13c15
|
491 492 |
* vma ranges that don't have struct page backing them or map I/O * devices directly cannot be handled by hmm_range_fault(). |
c2579c9c4
|
493 |
* |
d28c2c9a4
|
494 |
* If the vma does not allow read access, then assume that it does not |
c2579c9c4
|
495 496 |
* allow write access either. HMM does not support architectures that * allow write without read. |
a3eb13c15
|
497 498 499 |
* * If a fault is requested for an unsupported range then it is a hard * failure. |
d28c2c9a4
|
500 |
*/ |
a3eb13c15
|
501 |
if (hmm_range_need_fault(hmm_vma_walk, |
2733ea144
|
502 |
range->hmm_pfns + |
a3eb13c15
|
503 504 505 |
((start - range->start) >> PAGE_SHIFT), (end - start) >> PAGE_SHIFT, 0)) return -EFAULT; |
d28c2c9a4
|
506 |
|
a3eb13c15
|
507 |
hmm_pfns_fill(start, end, range, HMM_PFN_ERROR); |
d28c2c9a4
|
508 |
|
a3eb13c15
|
509 510 |
/* Skip this vma and continue processing the next vma. */ return 1; |
33cd47dcb
|
511 |
} |
7b86ac337
|
512 513 514 515 516 |
static const struct mm_walk_ops hmm_walk_ops = { .pud_entry = hmm_vma_walk_pud, .pmd_entry = hmm_vma_walk_pmd, .pte_hole = hmm_vma_walk_hole, .hugetlb_entry = hmm_vma_walk_hugetlb_entry, |
d28c2c9a4
|
517 |
.test_walk = hmm_vma_walk_test, |
7b86ac337
|
518 |
}; |
9a4903e49
|
519 520 |
/** * hmm_range_fault - try to fault some address in a virtual address range |
f970b977e
|
521 |
* @range: argument structure |
9a4903e49
|
522 |
* |
be957c886
|
523 |
* Returns 0 on success or one of the following error codes: |
73231612d
|
524 |
* |
9a4903e49
|
525 526 527 528 529 |
* -EINVAL: Invalid arguments or mm or virtual address is in an invalid vma * (e.g., device file vma). * -ENOMEM: Out of memory. * -EPERM: Invalid permission (e.g., asking for write and range is read * only). |
9a4903e49
|
530 531 |
* -EBUSY: The range has been invalidated and the caller needs to wait for * the invalidation to finish. |
f970b977e
|
532 533 |
* -EFAULT: A page was requested to be valid and could not be made valid * ie it has no backing VMA or it is illegal to access |
74eee180b
|
534 |
* |
f970b977e
|
535 536 |
* This is similar to get_user_pages(), except that it can read the page tables * without mutating them (ie causing faults). |
74eee180b
|
537 |
*/ |
be957c886
|
538 |
int hmm_range_fault(struct hmm_range *range) |
74eee180b
|
539 |
{ |
d28c2c9a4
|
540 541 542 |
struct hmm_vma_walk hmm_vma_walk = { .range = range, .last = range->start, |
d28c2c9a4
|
543 |
}; |
a22dd5064
|
544 |
struct mm_struct *mm = range->notifier->mm; |
74eee180b
|
545 |
int ret; |
42fc54140
|
546 |
mmap_assert_locked(mm); |
704f3f2cf
|
547 |
|
a3e0d41c2
|
548 549 |
do { /* If range is no longer valid force retry. */ |
a22dd5064
|
550 551 |
if (mmu_interval_check_retry(range->notifier, range->notifier_seq)) |
2bcbeaefd
|
552 |
return -EBUSY; |
d28c2c9a4
|
553 554 |
ret = walk_page_range(mm, hmm_vma_walk.last, range->end, &hmm_walk_ops, &hmm_vma_walk); |
be957c886
|
555 556 557 558 559 560 |
/* * When -EBUSY is returned the loop restarts with * hmm_vma_walk.last set to an address that has not been stored * in pfns. All entries < last in the pfn array are set to their * output, and all >= are still at their input values. */ |
d28c2c9a4
|
561 |
} while (ret == -EBUSY); |
be957c886
|
562 |
return ret; |
74eee180b
|
563 |
} |
73231612d
|
564 |
EXPORT_SYMBOL(hmm_range_fault); |