diff options
author | Benjamin Herrenschmidt <benh@kernel.crashing.org> | 2007-05-08 02:27:27 -0400 |
---|---|---|
committer | Paul Mackerras <paulus@samba.org> | 2007-05-09 02:35:00 -0400 |
commit | d0f13e3c20b6fb73ccb467bdca97fa7cf5a574cd (patch) | |
tree | a2de01a21dbb28449893102742e6b516a519c03e /arch/powerpc/mm/slice.c | |
parent | 16f1c746755836aa823658000493cdab8ce7b098 (diff) |
[POWERPC] Introduce address space "slices"
The basic issue is to be able to do what hugetlbfs does but with
different page sizes for some other special filesystems; more
specifically, my need is:
- Huge pages
- SPE local store mappings using 64K pages on a 4K base page size
kernel on Cell
- Some special 4K segments in 64K-page kernels for mapping a dodgy
type of powerpc-specific infiniband hardware that requires 4K MMU
mappings for various reasons I won't explain here.
The main issues are:
- To maintain/keep track of the page size per "segment" (as we can
only have one page size per segment on powerpc, which are 256MB
divisions of the address space).
- To make sure special mappings stay within their allotted
"segments" (including MAP_FIXED crap)
- To make sure everybody else doesn't mmap/brk/grow_stack into a
"segment" that is used for a special mapping
Some of the necessary mechanisms to handle that were present in the
hugetlbfs code, but mostly in ways not suitable for anything else.
The patch relies on some changes to the generic get_unmapped_area()
that just got merged. It still hijacks hugetlb callbacks here or
there as the generic code hasn't been entirely cleaned up yet but
that shouldn't be a problem.
So what is a slice ? Well, I re-used the mechanism used formerly by our
hugetlbfs implementation which divides the address space in
"meta-segments" which I called "slices". The division is done using
256MB slices below 4G, and 1T slices above. Thus the address space is
divided currently into 16 "low" slices and 16 "high" slices. (Special
case: high slice 0 is the area between 4G and 1T).
Doing so simplifies significantly the tracking of segments and avoids
having to keep track of all the 256MB segments in the address space.
While I used the "concepts" of hugetlbfs, I mostly re-implemented
everything in a more generic way and "ported" hugetlbfs to it.
Slices can have an associated page size, which is encoded in the mmu
context and used by the SLB miss handler to set the segment sizes. The
hash code currently doesn't care, it has a specific check for hugepages,
though I might add a mechanism to provide per-slice hash mapping
functions in the future.
The slice code provide a pair of "generic" get_unmapped_area() (bottomup
and topdown) functions that should work with any slice size. There is
some trickiness here so I would appreciate people to have a look at the
implementation of these and let me know if I got something wrong.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Diffstat (limited to 'arch/powerpc/mm/slice.c')
-rw-r--r-- | arch/powerpc/mm/slice.c | 633 |
1 files changed, 633 insertions, 0 deletions
diff --git a/arch/powerpc/mm/slice.c b/arch/powerpc/mm/slice.c new file mode 100644 index 000000000000..f833dba2a028 --- /dev/null +++ b/arch/powerpc/mm/slice.c | |||
@@ -0,0 +1,633 @@ | |||
1 | /* | ||
2 | * address space "slices" (meta-segments) support | ||
3 | * | ||
4 | * Copyright (C) 2007 Benjamin Herrenschmidt, IBM Corporation. | ||
5 | * | ||
6 | * Based on hugetlb implementation | ||
7 | * | ||
8 | * Copyright (C) 2003 David Gibson, IBM Corporation. | ||
9 | * | ||
10 | * This program is free software; you can redistribute it and/or modify | ||
11 | * it under the terms of the GNU General Public License as published by | ||
12 | * the Free Software Foundation; either version 2 of the License, or | ||
13 | * (at your option) any later version. | ||
14 | * | ||
15 | * This program is distributed in the hope that it will be useful, | ||
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
18 | * GNU General Public License for more details. | ||
19 | * | ||
20 | * You should have received a copy of the GNU General Public License | ||
21 | * along with this program; if not, write to the Free Software | ||
22 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
23 | */ | ||
24 | |||
25 | #undef DEBUG | ||
26 | |||
27 | #include <linux/kernel.h> | ||
28 | #include <linux/mm.h> | ||
29 | #include <linux/pagemap.h> | ||
30 | #include <linux/err.h> | ||
31 | #include <linux/spinlock.h> | ||
32 | #include <linux/module.h> | ||
33 | #include <asm/mman.h> | ||
34 | #include <asm/mmu.h> | ||
35 | #include <asm/spu.h> | ||
36 | |||
37 | static spinlock_t slice_convert_lock = SPIN_LOCK_UNLOCKED; | ||
38 | |||
39 | |||
40 | #ifdef DEBUG | ||
41 | int _slice_debug = 1; | ||
42 | |||
43 | static void slice_print_mask(const char *label, struct slice_mask mask) | ||
44 | { | ||
45 | char *p, buf[16 + 3 + 16 + 1]; | ||
46 | int i; | ||
47 | |||
48 | if (!_slice_debug) | ||
49 | return; | ||
50 | p = buf; | ||
51 | for (i = 0; i < SLICE_NUM_LOW; i++) | ||
52 | *(p++) = (mask.low_slices & (1 << i)) ? '1' : '0'; | ||
53 | *(p++) = ' '; | ||
54 | *(p++) = '-'; | ||
55 | *(p++) = ' '; | ||
56 | for (i = 0; i < SLICE_NUM_HIGH; i++) | ||
57 | *(p++) = (mask.high_slices & (1 << i)) ? '1' : '0'; | ||
58 | *(p++) = 0; | ||
59 | |||
60 | printk(KERN_DEBUG "%s:%s\n", label, buf); | ||
61 | } | ||
62 | |||
63 | #define slice_dbg(fmt...) do { if (_slice_debug) pr_debug(fmt); } while(0) | ||
64 | |||
65 | #else | ||
66 | |||
67 | static void slice_print_mask(const char *label, struct slice_mask mask) {} | ||
68 | #define slice_dbg(fmt...) | ||
69 | |||
70 | #endif | ||
71 | |||
72 | static struct slice_mask slice_range_to_mask(unsigned long start, | ||
73 | unsigned long len) | ||
74 | { | ||
75 | unsigned long end = start + len - 1; | ||
76 | struct slice_mask ret = { 0, 0 }; | ||
77 | |||
78 | if (start < SLICE_LOW_TOP) { | ||
79 | unsigned long mend = min(end, SLICE_LOW_TOP); | ||
80 | unsigned long mstart = min(start, SLICE_LOW_TOP); | ||
81 | |||
82 | ret.low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1)) | ||
83 | - (1u << GET_LOW_SLICE_INDEX(mstart)); | ||
84 | } | ||
85 | |||
86 | if ((start + len) > SLICE_LOW_TOP) | ||
87 | ret.high_slices = (1u << (GET_HIGH_SLICE_INDEX(end) + 1)) | ||
88 | - (1u << GET_HIGH_SLICE_INDEX(start)); | ||
89 | |||
90 | return ret; | ||
91 | } | ||
92 | |||
93 | static int slice_area_is_free(struct mm_struct *mm, unsigned long addr, | ||
94 | unsigned long len) | ||
95 | { | ||
96 | struct vm_area_struct *vma; | ||
97 | |||
98 | if ((mm->task_size - len) < addr) | ||
99 | return 0; | ||
100 | vma = find_vma(mm, addr); | ||
101 | return (!vma || (addr + len) <= vma->vm_start); | ||
102 | } | ||
103 | |||
104 | static int slice_low_has_vma(struct mm_struct *mm, unsigned long slice) | ||
105 | { | ||
106 | return !slice_area_is_free(mm, slice << SLICE_LOW_SHIFT, | ||
107 | 1ul << SLICE_LOW_SHIFT); | ||
108 | } | ||
109 | |||
110 | static int slice_high_has_vma(struct mm_struct *mm, unsigned long slice) | ||
111 | { | ||
112 | unsigned long start = slice << SLICE_HIGH_SHIFT; | ||
113 | unsigned long end = start + (1ul << SLICE_HIGH_SHIFT); | ||
114 | |||
115 | /* Hack, so that each addresses is controlled by exactly one | ||
116 | * of the high or low area bitmaps, the first high area starts | ||
117 | * at 4GB, not 0 */ | ||
118 | if (start == 0) | ||
119 | start = SLICE_LOW_TOP; | ||
120 | |||
121 | return !slice_area_is_free(mm, start, end - start); | ||
122 | } | ||
123 | |||
124 | static struct slice_mask slice_mask_for_free(struct mm_struct *mm) | ||
125 | { | ||
126 | struct slice_mask ret = { 0, 0 }; | ||
127 | unsigned long i; | ||
128 | |||
129 | for (i = 0; i < SLICE_NUM_LOW; i++) | ||
130 | if (!slice_low_has_vma(mm, i)) | ||
131 | ret.low_slices |= 1u << i; | ||
132 | |||
133 | if (mm->task_size <= SLICE_LOW_TOP) | ||
134 | return ret; | ||
135 | |||
136 | for (i = 0; i < SLICE_NUM_HIGH; i++) | ||
137 | if (!slice_high_has_vma(mm, i)) | ||
138 | ret.high_slices |= 1u << i; | ||
139 | |||
140 | return ret; | ||
141 | } | ||
142 | |||
143 | static struct slice_mask slice_mask_for_size(struct mm_struct *mm, int psize) | ||
144 | { | ||
145 | struct slice_mask ret = { 0, 0 }; | ||
146 | unsigned long i; | ||
147 | u64 psizes; | ||
148 | |||
149 | psizes = mm->context.low_slices_psize; | ||
150 | for (i = 0; i < SLICE_NUM_LOW; i++) | ||
151 | if (((psizes >> (i * 4)) & 0xf) == psize) | ||
152 | ret.low_slices |= 1u << i; | ||
153 | |||
154 | psizes = mm->context.high_slices_psize; | ||
155 | for (i = 0; i < SLICE_NUM_HIGH; i++) | ||
156 | if (((psizes >> (i * 4)) & 0xf) == psize) | ||
157 | ret.high_slices |= 1u << i; | ||
158 | |||
159 | return ret; | ||
160 | } | ||
161 | |||
162 | static int slice_check_fit(struct slice_mask mask, struct slice_mask available) | ||
163 | { | ||
164 | return (mask.low_slices & available.low_slices) == mask.low_slices && | ||
165 | (mask.high_slices & available.high_slices) == mask.high_slices; | ||
166 | } | ||
167 | |||
168 | static void slice_flush_segments(void *parm) | ||
169 | { | ||
170 | struct mm_struct *mm = parm; | ||
171 | unsigned long flags; | ||
172 | |||
173 | if (mm != current->active_mm) | ||
174 | return; | ||
175 | |||
176 | /* update the paca copy of the context struct */ | ||
177 | get_paca()->context = current->active_mm->context; | ||
178 | |||
179 | local_irq_save(flags); | ||
180 | slb_flush_and_rebolt(); | ||
181 | local_irq_restore(flags); | ||
182 | } | ||
183 | |||
184 | static void slice_convert(struct mm_struct *mm, struct slice_mask mask, int psize) | ||
185 | { | ||
186 | /* Write the new slice psize bits */ | ||
187 | u64 lpsizes, hpsizes; | ||
188 | unsigned long i, flags; | ||
189 | |||
190 | slice_dbg("slice_convert(mm=%p, psize=%d)\n", mm, psize); | ||
191 | slice_print_mask(" mask", mask); | ||
192 | |||
193 | /* We need to use a spinlock here to protect against | ||
194 | * concurrent 64k -> 4k demotion ... | ||
195 | */ | ||
196 | spin_lock_irqsave(&slice_convert_lock, flags); | ||
197 | |||
198 | lpsizes = mm->context.low_slices_psize; | ||
199 | for (i = 0; i < SLICE_NUM_LOW; i++) | ||
200 | if (mask.low_slices & (1u << i)) | ||
201 | lpsizes = (lpsizes & ~(0xful << (i * 4))) | | ||
202 | (((unsigned long)psize) << (i * 4)); | ||
203 | |||
204 | hpsizes = mm->context.high_slices_psize; | ||
205 | for (i = 0; i < SLICE_NUM_HIGH; i++) | ||
206 | if (mask.high_slices & (1u << i)) | ||
207 | hpsizes = (hpsizes & ~(0xful << (i * 4))) | | ||
208 | (((unsigned long)psize) << (i * 4)); | ||
209 | |||
210 | mm->context.low_slices_psize = lpsizes; | ||
211 | mm->context.high_slices_psize = hpsizes; | ||
212 | |||
213 | slice_dbg(" lsps=%lx, hsps=%lx\n", | ||
214 | mm->context.low_slices_psize, | ||
215 | mm->context.high_slices_psize); | ||
216 | |||
217 | spin_unlock_irqrestore(&slice_convert_lock, flags); | ||
218 | mb(); | ||
219 | |||
220 | /* XXX this is sub-optimal but will do for now */ | ||
221 | on_each_cpu(slice_flush_segments, mm, 0, 1); | ||
222 | #ifdef CONFIG_SPU_BASE | ||
223 | spu_flush_all_slbs(mm); | ||
224 | #endif | ||
225 | } | ||
226 | |||
227 | static unsigned long slice_find_area_bottomup(struct mm_struct *mm, | ||
228 | unsigned long len, | ||
229 | struct slice_mask available, | ||
230 | int psize, int use_cache) | ||
231 | { | ||
232 | struct vm_area_struct *vma; | ||
233 | unsigned long start_addr, addr; | ||
234 | struct slice_mask mask; | ||
235 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); | ||
236 | |||
237 | if (use_cache) { | ||
238 | if (len <= mm->cached_hole_size) { | ||
239 | start_addr = addr = TASK_UNMAPPED_BASE; | ||
240 | mm->cached_hole_size = 0; | ||
241 | } else | ||
242 | start_addr = addr = mm->free_area_cache; | ||
243 | } else | ||
244 | start_addr = addr = TASK_UNMAPPED_BASE; | ||
245 | |||
246 | full_search: | ||
247 | for (;;) { | ||
248 | addr = _ALIGN_UP(addr, 1ul << pshift); | ||
249 | if ((TASK_SIZE - len) < addr) | ||
250 | break; | ||
251 | vma = find_vma(mm, addr); | ||
252 | BUG_ON(vma && (addr >= vma->vm_end)); | ||
253 | |||
254 | mask = slice_range_to_mask(addr, len); | ||
255 | if (!slice_check_fit(mask, available)) { | ||
256 | if (addr < SLICE_LOW_TOP) | ||
257 | addr = _ALIGN_UP(addr + 1, 1ul << SLICE_LOW_SHIFT); | ||
258 | else | ||
259 | addr = _ALIGN_UP(addr + 1, 1ul << SLICE_HIGH_SHIFT); | ||
260 | continue; | ||
261 | } | ||
262 | if (!vma || addr + len <= vma->vm_start) { | ||
263 | /* | ||
264 | * Remember the place where we stopped the search: | ||
265 | */ | ||
266 | if (use_cache) | ||
267 | mm->free_area_cache = addr + len; | ||
268 | return addr; | ||
269 | } | ||
270 | if (use_cache && (addr + mm->cached_hole_size) < vma->vm_start) | ||
271 | mm->cached_hole_size = vma->vm_start - addr; | ||
272 | addr = vma->vm_end; | ||
273 | } | ||
274 | |||
275 | /* Make sure we didn't miss any holes */ | ||
276 | if (use_cache && start_addr != TASK_UNMAPPED_BASE) { | ||
277 | start_addr = addr = TASK_UNMAPPED_BASE; | ||
278 | mm->cached_hole_size = 0; | ||
279 | goto full_search; | ||
280 | } | ||
281 | return -ENOMEM; | ||
282 | } | ||
283 | |||
284 | static unsigned long slice_find_area_topdown(struct mm_struct *mm, | ||
285 | unsigned long len, | ||
286 | struct slice_mask available, | ||
287 | int psize, int use_cache) | ||
288 | { | ||
289 | struct vm_area_struct *vma; | ||
290 | unsigned long addr; | ||
291 | struct slice_mask mask; | ||
292 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); | ||
293 | |||
294 | /* check if free_area_cache is useful for us */ | ||
295 | if (use_cache) { | ||
296 | if (len <= mm->cached_hole_size) { | ||
297 | mm->cached_hole_size = 0; | ||
298 | mm->free_area_cache = mm->mmap_base; | ||
299 | } | ||
300 | |||
301 | /* either no address requested or can't fit in requested | ||
302 | * address hole | ||
303 | */ | ||
304 | addr = mm->free_area_cache; | ||
305 | |||
306 | /* make sure it can fit in the remaining address space */ | ||
307 | if (addr > len) { | ||
308 | addr = _ALIGN_DOWN(addr - len, 1ul << pshift); | ||
309 | mask = slice_range_to_mask(addr, len); | ||
310 | if (slice_check_fit(mask, available) && | ||
311 | slice_area_is_free(mm, addr, len)) | ||
312 | /* remember the address as a hint for | ||
313 | * next time | ||
314 | */ | ||
315 | return (mm->free_area_cache = addr); | ||
316 | } | ||
317 | } | ||
318 | |||
319 | addr = mm->mmap_base; | ||
320 | while (addr > len) { | ||
321 | /* Go down by chunk size */ | ||
322 | addr = _ALIGN_DOWN(addr - len, 1ul << pshift); | ||
323 | |||
324 | /* Check for hit with different page size */ | ||
325 | mask = slice_range_to_mask(addr, len); | ||
326 | if (!slice_check_fit(mask, available)) { | ||
327 | if (addr < SLICE_LOW_TOP) | ||
328 | addr = _ALIGN_DOWN(addr, 1ul << SLICE_LOW_SHIFT); | ||
329 | else if (addr < (1ul << SLICE_HIGH_SHIFT)) | ||
330 | addr = SLICE_LOW_TOP; | ||
331 | else | ||
332 | addr = _ALIGN_DOWN(addr, 1ul << SLICE_HIGH_SHIFT); | ||
333 | continue; | ||
334 | } | ||
335 | |||
336 | /* | ||
337 | * Lookup failure means no vma is above this address, | ||
338 | * else if new region fits below vma->vm_start, | ||
339 | * return with success: | ||
340 | */ | ||
341 | vma = find_vma(mm, addr); | ||
342 | if (!vma || (addr + len) <= vma->vm_start) { | ||
343 | /* remember the address as a hint for next time */ | ||
344 | if (use_cache) | ||
345 | mm->free_area_cache = addr; | ||
346 | return addr; | ||
347 | } | ||
348 | |||
349 | /* remember the largest hole we saw so far */ | ||
350 | if (use_cache && (addr + mm->cached_hole_size) < vma->vm_start) | ||
351 | mm->cached_hole_size = vma->vm_start - addr; | ||
352 | |||
353 | /* try just below the current vma->vm_start */ | ||
354 | addr = vma->vm_start; | ||
355 | } | ||
356 | |||
357 | /* | ||
358 | * A failed mmap() very likely causes application failure, | ||
359 | * so fall back to the bottom-up function here. This scenario | ||
360 | * can happen with large stack limits and large mmap() | ||
361 | * allocations. | ||
362 | */ | ||
363 | addr = slice_find_area_bottomup(mm, len, available, psize, 0); | ||
364 | |||
365 | /* | ||
366 | * Restore the topdown base: | ||
367 | */ | ||
368 | if (use_cache) { | ||
369 | mm->free_area_cache = mm->mmap_base; | ||
370 | mm->cached_hole_size = ~0UL; | ||
371 | } | ||
372 | |||
373 | return addr; | ||
374 | } | ||
375 | |||
376 | |||
377 | static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len, | ||
378 | struct slice_mask mask, int psize, | ||
379 | int topdown, int use_cache) | ||
380 | { | ||
381 | if (topdown) | ||
382 | return slice_find_area_topdown(mm, len, mask, psize, use_cache); | ||
383 | else | ||
384 | return slice_find_area_bottomup(mm, len, mask, psize, use_cache); | ||
385 | } | ||
386 | |||
387 | unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len, | ||
388 | unsigned long flags, unsigned int psize, | ||
389 | int topdown, int use_cache) | ||
390 | { | ||
391 | struct slice_mask mask; | ||
392 | struct slice_mask good_mask; | ||
393 | struct slice_mask potential_mask = {0,0} /* silence stupid warning */; | ||
394 | int pmask_set = 0; | ||
395 | int fixed = (flags & MAP_FIXED); | ||
396 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); | ||
397 | struct mm_struct *mm = current->mm; | ||
398 | |||
399 | /* Sanity checks */ | ||
400 | BUG_ON(mm->task_size == 0); | ||
401 | |||
402 | slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize); | ||
403 | slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d, use_cache=%d\n", | ||
404 | addr, len, flags, topdown, use_cache); | ||
405 | |||
406 | if (len > mm->task_size) | ||
407 | return -ENOMEM; | ||
408 | if (fixed && (addr & ((1ul << pshift) - 1))) | ||
409 | return -EINVAL; | ||
410 | if (fixed && addr > (mm->task_size - len)) | ||
411 | return -EINVAL; | ||
412 | |||
413 | /* If hint, make sure it matches our alignment restrictions */ | ||
414 | if (!fixed && addr) { | ||
415 | addr = _ALIGN_UP(addr, 1ul << pshift); | ||
416 | slice_dbg(" aligned addr=%lx\n", addr); | ||
417 | } | ||
418 | |||
419 | /* First makeup a "good" mask of slices that have the right size | ||
420 | * already | ||
421 | */ | ||
422 | good_mask = slice_mask_for_size(mm, psize); | ||
423 | slice_print_mask(" good_mask", good_mask); | ||
424 | |||
425 | /* First check hint if it's valid or if we have MAP_FIXED */ | ||
426 | if ((addr != 0 || fixed) && (mm->task_size - len) >= addr) { | ||
427 | |||
428 | /* Don't bother with hint if it overlaps a VMA */ | ||
429 | if (!fixed && !slice_area_is_free(mm, addr, len)) | ||
430 | goto search; | ||
431 | |||
432 | /* Build a mask for the requested range */ | ||
433 | mask = slice_range_to_mask(addr, len); | ||
434 | slice_print_mask(" mask", mask); | ||
435 | |||
436 | /* Check if we fit in the good mask. If we do, we just return, | ||
437 | * nothing else to do | ||
438 | */ | ||
439 | if (slice_check_fit(mask, good_mask)) { | ||
440 | slice_dbg(" fits good !\n"); | ||
441 | return addr; | ||
442 | } | ||
443 | |||
444 | /* We don't fit in the good mask, check what other slices are | ||
445 | * empty and thus can be converted | ||
446 | */ | ||
447 | potential_mask = slice_mask_for_free(mm); | ||
448 | potential_mask.low_slices |= good_mask.low_slices; | ||
449 | potential_mask.high_slices |= good_mask.high_slices; | ||
450 | pmask_set = 1; | ||
451 | slice_print_mask(" potential", potential_mask); | ||
452 | if (slice_check_fit(mask, potential_mask)) { | ||
453 | slice_dbg(" fits potential !\n"); | ||
454 | goto convert; | ||
455 | } | ||
456 | } | ||
457 | |||
458 | /* If we have MAP_FIXED and failed the above step, then error out */ | ||
459 | if (fixed) | ||
460 | return -EBUSY; | ||
461 | |||
462 | search: | ||
463 | slice_dbg(" search...\n"); | ||
464 | |||
465 | /* Now let's see if we can find something in the existing slices | ||
466 | * for that size | ||
467 | */ | ||
468 | addr = slice_find_area(mm, len, good_mask, psize, topdown, use_cache); | ||
469 | if (addr != -ENOMEM) { | ||
470 | /* Found within the good mask, we don't have to setup, | ||
471 | * we thus return directly | ||
472 | */ | ||
473 | slice_dbg(" found area at 0x%lx\n", addr); | ||
474 | return addr; | ||
475 | } | ||
476 | |||
477 | /* Won't fit, check what can be converted */ | ||
478 | if (!pmask_set) { | ||
479 | potential_mask = slice_mask_for_free(mm); | ||
480 | potential_mask.low_slices |= good_mask.low_slices; | ||
481 | potential_mask.high_slices |= good_mask.high_slices; | ||
482 | pmask_set = 1; | ||
483 | slice_print_mask(" potential", potential_mask); | ||
484 | } | ||
485 | |||
486 | /* Now let's see if we can find something in the existing slices | ||
487 | * for that size | ||
488 | */ | ||
489 | addr = slice_find_area(mm, len, potential_mask, psize, topdown, | ||
490 | use_cache); | ||
491 | if (addr == -ENOMEM) | ||
492 | return -ENOMEM; | ||
493 | |||
494 | mask = slice_range_to_mask(addr, len); | ||
495 | slice_dbg(" found potential area at 0x%lx\n", addr); | ||
496 | slice_print_mask(" mask", mask); | ||
497 | |||
498 | convert: | ||
499 | slice_convert(mm, mask, psize); | ||
500 | return addr; | ||
501 | |||
502 | } | ||
503 | EXPORT_SYMBOL_GPL(slice_get_unmapped_area); | ||
504 | |||
505 | unsigned long arch_get_unmapped_area(struct file *filp, | ||
506 | unsigned long addr, | ||
507 | unsigned long len, | ||
508 | unsigned long pgoff, | ||
509 | unsigned long flags) | ||
510 | { | ||
511 | return slice_get_unmapped_area(addr, len, flags, | ||
512 | current->mm->context.user_psize, | ||
513 | 0, 1); | ||
514 | } | ||
515 | |||
516 | unsigned long arch_get_unmapped_area_topdown(struct file *filp, | ||
517 | const unsigned long addr0, | ||
518 | const unsigned long len, | ||
519 | const unsigned long pgoff, | ||
520 | const unsigned long flags) | ||
521 | { | ||
522 | return slice_get_unmapped_area(addr0, len, flags, | ||
523 | current->mm->context.user_psize, | ||
524 | 1, 1); | ||
525 | } | ||
526 | |||
527 | unsigned int get_slice_psize(struct mm_struct *mm, unsigned long addr) | ||
528 | { | ||
529 | u64 psizes; | ||
530 | int index; | ||
531 | |||
532 | if (addr < SLICE_LOW_TOP) { | ||
533 | psizes = mm->context.low_slices_psize; | ||
534 | index = GET_LOW_SLICE_INDEX(addr); | ||
535 | } else { | ||
536 | psizes = mm->context.high_slices_psize; | ||
537 | index = GET_HIGH_SLICE_INDEX(addr); | ||
538 | } | ||
539 | |||
540 | return (psizes >> (index * 4)) & 0xf; | ||
541 | } | ||
542 | EXPORT_SYMBOL_GPL(get_slice_psize); | ||
543 | |||
544 | /* | ||
545 | * This is called by hash_page when it needs to do a lazy conversion of | ||
546 | * an address space from real 64K pages to combo 4K pages (typically | ||
547 | * when hitting a non cacheable mapping on a processor or hypervisor | ||
548 | * that won't allow them for 64K pages). | ||
549 | * | ||
550 | * This is also called in init_new_context() to change back the user | ||
551 | * psize from whatever the parent context had it set to | ||
552 | * | ||
553 | * This function will only change the content of the {low,high)_slice_psize | ||
554 | * masks, it will not flush SLBs as this shall be handled lazily by the | ||
555 | * caller. | ||
556 | */ | ||
557 | void slice_set_user_psize(struct mm_struct *mm, unsigned int psize) | ||
558 | { | ||
559 | unsigned long flags, lpsizes, hpsizes; | ||
560 | unsigned int old_psize; | ||
561 | int i; | ||
562 | |||
563 | slice_dbg("slice_set_user_psize(mm=%p, psize=%d)\n", mm, psize); | ||
564 | |||
565 | spin_lock_irqsave(&slice_convert_lock, flags); | ||
566 | |||
567 | old_psize = mm->context.user_psize; | ||
568 | slice_dbg(" old_psize=%d\n", old_psize); | ||
569 | if (old_psize == psize) | ||
570 | goto bail; | ||
571 | |||
572 | mm->context.user_psize = psize; | ||
573 | wmb(); | ||
574 | |||
575 | lpsizes = mm->context.low_slices_psize; | ||
576 | for (i = 0; i < SLICE_NUM_LOW; i++) | ||
577 | if (((lpsizes >> (i * 4)) & 0xf) == old_psize) | ||
578 | lpsizes = (lpsizes & ~(0xful << (i * 4))) | | ||
579 | (((unsigned long)psize) << (i * 4)); | ||
580 | |||
581 | hpsizes = mm->context.high_slices_psize; | ||
582 | for (i = 0; i < SLICE_NUM_HIGH; i++) | ||
583 | if (((hpsizes >> (i * 4)) & 0xf) == old_psize) | ||
584 | hpsizes = (hpsizes & ~(0xful << (i * 4))) | | ||
585 | (((unsigned long)psize) << (i * 4)); | ||
586 | |||
587 | mm->context.low_slices_psize = lpsizes; | ||
588 | mm->context.high_slices_psize = hpsizes; | ||
589 | |||
590 | slice_dbg(" lsps=%lx, hsps=%lx\n", | ||
591 | mm->context.low_slices_psize, | ||
592 | mm->context.high_slices_psize); | ||
593 | |||
594 | bail: | ||
595 | spin_unlock_irqrestore(&slice_convert_lock, flags); | ||
596 | } | ||
597 | |||
598 | /* | ||
599 | * is_hugepage_only_range() is used by generic code to verify wether | ||
600 | * a normal mmap mapping (non hugetlbfs) is valid on a given area. | ||
601 | * | ||
602 | * until the generic code provides a more generic hook and/or starts | ||
603 | * calling arch get_unmapped_area for MAP_FIXED (which our implementation | ||
604 | * here knows how to deal with), we hijack it to keep standard mappings | ||
605 | * away from us. | ||
606 | * | ||
607 | * because of that generic code limitation, MAP_FIXED mapping cannot | ||
608 | * "convert" back a slice with no VMAs to the standard page size, only | ||
609 | * get_unmapped_area() can. It would be possible to fix it here but I | ||
610 | * prefer working on fixing the generic code instead. | ||
611 | * | ||
612 | * WARNING: This will not work if hugetlbfs isn't enabled since the | ||
613 | * generic code will redefine that function as 0 in that. This is ok | ||
614 | * for now as we only use slices with hugetlbfs enabled. This should | ||
615 | * be fixed as the generic code gets fixed. | ||
616 | */ | ||
617 | int is_hugepage_only_range(struct mm_struct *mm, unsigned long addr, | ||
618 | unsigned long len) | ||
619 | { | ||
620 | struct slice_mask mask, available; | ||
621 | |||
622 | mask = slice_range_to_mask(addr, len); | ||
623 | available = slice_mask_for_size(mm, mm->context.user_psize); | ||
624 | |||
625 | #if 0 /* too verbose */ | ||
626 | slice_dbg("is_hugepage_only_range(mm=%p, addr=%lx, len=%lx)\n", | ||
627 | mm, addr, len); | ||
628 | slice_print_mask(" mask", mask); | ||
629 | slice_print_mask(" available", available); | ||
630 | #endif | ||
631 | return !slice_check_fit(mask, available); | ||
632 | } | ||
633 | |||