Commit | Line | Data |
---|---|---|
5bc20fc5 DM |
1 | /* |
2 | * Xen implementation for transcendent memory (tmem) | |
3 | * | |
afec6e04 | 4 | * Copyright (C) 2009-2011 Oracle Corp. All rights reserved. |
5bc20fc5 DM |
5 | * Author: Dan Magenheimer |
6 | */ | |
7 | ||
8 | #include <linux/kernel.h> | |
9 | #include <linux/types.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/cleancache.h> | |
13 | ||
afec6e04 DM |
14 | /* temporary ifdef until include/linux/frontswap.h is upstream */ |
15 | #ifdef CONFIG_FRONTSWAP | |
16 | #include <linux/frontswap.h> | |
17 | #endif | |
18 | ||
5bc20fc5 DM |
19 | #include <xen/xen.h> |
20 | #include <xen/interface/xen.h> | |
21 | #include <asm/xen/hypercall.h> | |
22 | #include <asm/xen/page.h> | |
23 | #include <asm/xen/hypervisor.h> | |
24 | ||
25 | #define TMEM_CONTROL 0 | |
26 | #define TMEM_NEW_POOL 1 | |
27 | #define TMEM_DESTROY_POOL 2 | |
28 | #define TMEM_NEW_PAGE 3 | |
29 | #define TMEM_PUT_PAGE 4 | |
30 | #define TMEM_GET_PAGE 5 | |
31 | #define TMEM_FLUSH_PAGE 6 | |
32 | #define TMEM_FLUSH_OBJECT 7 | |
33 | #define TMEM_READ 8 | |
34 | #define TMEM_WRITE 9 | |
35 | #define TMEM_XCHG 10 | |
36 | ||
37 | /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */ | |
38 | #define TMEM_POOL_PERSIST 1 | |
39 | #define TMEM_POOL_SHARED 2 | |
40 | #define TMEM_POOL_PAGESIZE_SHIFT 4 | |
41 | #define TMEM_VERSION_SHIFT 24 | |
42 | ||
43 | ||
44 | struct tmem_pool_uuid { | |
45 | u64 uuid_lo; | |
46 | u64 uuid_hi; | |
47 | }; | |
48 | ||
49 | struct tmem_oid { | |
50 | u64 oid[3]; | |
51 | }; | |
52 | ||
53 | #define TMEM_POOL_PRIVATE_UUID { 0, 0 } | |
54 | ||
55 | /* flags for tmem_ops.new_pool */ | |
56 | #define TMEM_POOL_PERSIST 1 | |
57 | #define TMEM_POOL_SHARED 2 | |
58 | ||
59 | /* xen tmem foundation ops/hypercalls */ | |
60 | ||
61 | static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid, | |
62 | u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len) | |
63 | { | |
64 | struct tmem_op op; | |
65 | int rc = 0; | |
66 | ||
67 | op.cmd = tmem_cmd; | |
68 | op.pool_id = tmem_pool; | |
69 | op.u.gen.oid[0] = oid.oid[0]; | |
70 | op.u.gen.oid[1] = oid.oid[1]; | |
71 | op.u.gen.oid[2] = oid.oid[2]; | |
72 | op.u.gen.index = index; | |
73 | op.u.gen.tmem_offset = tmem_offset; | |
74 | op.u.gen.pfn_offset = pfn_offset; | |
75 | op.u.gen.len = len; | |
76 | set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn); | |
77 | rc = HYPERVISOR_tmem_op(&op); | |
78 | return rc; | |
79 | } | |
80 | ||
81 | static int xen_tmem_new_pool(struct tmem_pool_uuid uuid, | |
82 | u32 flags, unsigned long pagesize) | |
83 | { | |
84 | struct tmem_op op; | |
85 | int rc = 0, pageshift; | |
86 | ||
87 | for (pageshift = 0; pagesize != 1; pageshift++) | |
88 | pagesize >>= 1; | |
89 | flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT; | |
90 | flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT; | |
91 | op.cmd = TMEM_NEW_POOL; | |
92 | op.u.new.uuid[0] = uuid.uuid_lo; | |
93 | op.u.new.uuid[1] = uuid.uuid_hi; | |
94 | op.u.new.flags = flags; | |
95 | rc = HYPERVISOR_tmem_op(&op); | |
96 | return rc; | |
97 | } | |
98 | ||
99 | /* xen generic tmem ops */ | |
100 | ||
101 | static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid, | |
102 | u32 index, unsigned long pfn) | |
103 | { | |
104 | unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn; | |
105 | ||
106 | return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index, | |
107 | gmfn, 0, 0, 0); | |
108 | } | |
109 | ||
110 | static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid, | |
111 | u32 index, unsigned long pfn) | |
112 | { | |
113 | unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn; | |
114 | ||
115 | return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index, | |
116 | gmfn, 0, 0, 0); | |
117 | } | |
118 | ||
119 | static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index) | |
120 | { | |
121 | return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index, | |
122 | 0, 0, 0, 0); | |
123 | } | |
124 | ||
125 | static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid) | |
126 | { | |
127 | return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0); | |
128 | } | |
129 | ||
8e6f7c23 | 130 | bool __read_mostly tmem_enabled = false; |
5bc20fc5 DM |
131 | |
132 | static int __init enable_tmem(char *s) | |
133 | { | |
8e6f7c23 | 134 | tmem_enabled = true; |
5bc20fc5 DM |
135 | return 1; |
136 | } | |
5bc20fc5 DM |
137 | __setup("tmem", enable_tmem); |
138 | ||
afec6e04 DM |
139 | #ifdef CONFIG_CLEANCACHE |
140 | static int xen_tmem_destroy_pool(u32 pool_id) | |
141 | { | |
142 | struct tmem_oid oid = { { 0 } }; | |
143 | ||
144 | return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0); | |
145 | } | |
146 | ||
5bc20fc5 DM |
147 | /* cleancache ops */ |
148 | ||
149 | static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key, | |
150 | pgoff_t index, struct page *page) | |
151 | { | |
152 | u32 ind = (u32) index; | |
153 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
154 | unsigned long pfn = page_to_pfn(page); | |
155 | ||
156 | if (pool < 0) | |
157 | return; | |
158 | if (ind != index) | |
159 | return; | |
160 | mb(); /* ensure page is quiescent; tmem may address it with an alias */ | |
161 | (void)xen_tmem_put_page((u32)pool, oid, ind, pfn); | |
162 | } | |
163 | ||
164 | static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key, | |
165 | pgoff_t index, struct page *page) | |
166 | { | |
167 | u32 ind = (u32) index; | |
168 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
169 | unsigned long pfn = page_to_pfn(page); | |
170 | int ret; | |
171 | ||
172 | /* translate return values to linux semantics */ | |
173 | if (pool < 0) | |
174 | return -1; | |
175 | if (ind != index) | |
176 | return -1; | |
177 | ret = xen_tmem_get_page((u32)pool, oid, ind, pfn); | |
178 | if (ret == 1) | |
179 | return 0; | |
180 | else | |
181 | return -1; | |
182 | } | |
183 | ||
184 | static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key, | |
185 | pgoff_t index) | |
186 | { | |
187 | u32 ind = (u32) index; | |
188 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
189 | ||
190 | if (pool < 0) | |
191 | return; | |
192 | if (ind != index) | |
193 | return; | |
194 | (void)xen_tmem_flush_page((u32)pool, oid, ind); | |
195 | } | |
196 | ||
197 | static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key) | |
198 | { | |
199 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
200 | ||
201 | if (pool < 0) | |
202 | return; | |
203 | (void)xen_tmem_flush_object((u32)pool, oid); | |
204 | } | |
205 | ||
206 | static void tmem_cleancache_flush_fs(int pool) | |
207 | { | |
208 | if (pool < 0) | |
209 | return; | |
210 | (void)xen_tmem_destroy_pool((u32)pool); | |
211 | } | |
212 | ||
213 | static int tmem_cleancache_init_fs(size_t pagesize) | |
214 | { | |
215 | struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID; | |
216 | ||
217 | return xen_tmem_new_pool(uuid_private, 0, pagesize); | |
218 | } | |
219 | ||
220 | static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize) | |
221 | { | |
222 | struct tmem_pool_uuid shared_uuid; | |
223 | ||
224 | shared_uuid.uuid_lo = *(u64 *)uuid; | |
225 | shared_uuid.uuid_hi = *(u64 *)(&uuid[8]); | |
226 | return xen_tmem_new_pool(shared_uuid, TMEM_POOL_SHARED, pagesize); | |
227 | } | |
228 | ||
8e6f7c23 | 229 | static bool __initdata use_cleancache = true; |
5bc20fc5 DM |
230 | |
231 | static int __init no_cleancache(char *s) | |
232 | { | |
8e6f7c23 | 233 | use_cleancache = false; |
5bc20fc5 DM |
234 | return 1; |
235 | } | |
5bc20fc5 DM |
236 | __setup("nocleancache", no_cleancache); |
237 | ||
8e6f7c23 | 238 | static struct cleancache_ops __initdata tmem_cleancache_ops = { |
5bc20fc5 DM |
239 | .put_page = tmem_cleancache_put_page, |
240 | .get_page = tmem_cleancache_get_page, | |
91c6cc9b DM |
241 | .invalidate_page = tmem_cleancache_flush_page, |
242 | .invalidate_inode = tmem_cleancache_flush_inode, | |
243 | .invalidate_fs = tmem_cleancache_flush_fs, | |
5bc20fc5 DM |
244 | .init_shared_fs = tmem_cleancache_init_shared_fs, |
245 | .init_fs = tmem_cleancache_init_fs | |
246 | }; | |
afec6e04 | 247 | #endif |
5bc20fc5 | 248 | |
afec6e04 DM |
249 | #ifdef CONFIG_FRONTSWAP |
250 | /* frontswap tmem operations */ | |
251 | ||
252 | /* a single tmem poolid is used for all frontswap "types" (swapfiles) */ | |
253 | static int tmem_frontswap_poolid; | |
254 | ||
255 | /* | |
256 | * Swizzling increases objects per swaptype, increasing tmem concurrency | |
257 | * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS | |
258 | */ | |
259 | #define SWIZ_BITS 4 | |
260 | #define SWIZ_MASK ((1 << SWIZ_BITS) - 1) | |
261 | #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK)) | |
262 | #define iswiz(_ind) (_ind >> SWIZ_BITS) | |
263 | ||
264 | static inline struct tmem_oid oswiz(unsigned type, u32 ind) | |
5bc20fc5 | 265 | { |
afec6e04 DM |
266 | struct tmem_oid oid = { .oid = { 0 } }; |
267 | oid.oid[0] = _oswiz(type, ind); | |
268 | return oid; | |
269 | } | |
5bc20fc5 | 270 | |
afec6e04 DM |
271 | /* returns 0 if the page was successfully put into frontswap, -1 if not */ |
272 | static int tmem_frontswap_put_page(unsigned type, pgoff_t offset, | |
273 | struct page *page) | |
274 | { | |
275 | u64 ind64 = (u64)offset; | |
276 | u32 ind = (u32)offset; | |
277 | unsigned long pfn = page_to_pfn(page); | |
278 | int pool = tmem_frontswap_poolid; | |
279 | int ret; | |
280 | ||
281 | if (pool < 0) | |
282 | return -1; | |
283 | if (ind64 != ind) | |
284 | return -1; | |
285 | mb(); /* ensure page is quiescent; tmem may address it with an alias */ | |
286 | ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn); | |
287 | /* translate Xen tmem return values to linux semantics */ | |
288 | if (ret == 1) | |
289 | return 0; | |
290 | else | |
291 | return -1; | |
292 | } | |
293 | ||
294 | /* | |
295 | * returns 0 if the page was successfully gotten from frontswap, -1 if | |
296 | * was not present (should never happen!) | |
297 | */ | |
298 | static int tmem_frontswap_get_page(unsigned type, pgoff_t offset, | |
299 | struct page *page) | |
300 | { | |
301 | u64 ind64 = (u64)offset; | |
302 | u32 ind = (u32)offset; | |
303 | unsigned long pfn = page_to_pfn(page); | |
304 | int pool = tmem_frontswap_poolid; | |
305 | int ret; | |
306 | ||
307 | if (pool < 0) | |
308 | return -1; | |
309 | if (ind64 != ind) | |
310 | return -1; | |
311 | ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn); | |
312 | /* translate Xen tmem return values to linux semantics */ | |
313 | if (ret == 1) | |
314 | return 0; | |
315 | else | |
316 | return -1; | |
317 | } | |
318 | ||
319 | /* flush a single page from frontswap */ | |
320 | static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset) | |
321 | { | |
322 | u64 ind64 = (u64)offset; | |
323 | u32 ind = (u32)offset; | |
324 | int pool = tmem_frontswap_poolid; | |
325 | ||
326 | if (pool < 0) | |
327 | return; | |
328 | if (ind64 != ind) | |
329 | return; | |
330 | (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind)); | |
331 | } | |
332 | ||
333 | /* flush all pages from the passed swaptype */ | |
334 | static void tmem_frontswap_flush_area(unsigned type) | |
335 | { | |
336 | int pool = tmem_frontswap_poolid; | |
337 | int ind; | |
338 | ||
339 | if (pool < 0) | |
340 | return; | |
341 | for (ind = SWIZ_MASK; ind >= 0; ind--) | |
342 | (void)xen_tmem_flush_object(pool, oswiz(type, ind)); | |
343 | } | |
344 | ||
345 | static void tmem_frontswap_init(unsigned ignored) | |
346 | { | |
347 | struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID; | |
348 | ||
349 | /* a single tmem poolid is used for all frontswap "types" (swapfiles) */ | |
350 | if (tmem_frontswap_poolid < 0) | |
351 | tmem_frontswap_poolid = | |
352 | xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE); | |
353 | } | |
354 | ||
8e6f7c23 | 355 | static bool __initdata use_frontswap = true; |
afec6e04 DM |
356 | |
357 | static int __init no_frontswap(char *s) | |
358 | { | |
8e6f7c23 | 359 | use_frontswap = false; |
afec6e04 DM |
360 | return 1; |
361 | } | |
afec6e04 DM |
362 | __setup("nofrontswap", no_frontswap); |
363 | ||
8e6f7c23 | 364 | static struct frontswap_ops __initdata tmem_frontswap_ops = { |
afec6e04 DM |
365 | .put_page = tmem_frontswap_put_page, |
366 | .get_page = tmem_frontswap_get_page, | |
91c6cc9b DM |
367 | .invalidate_page = tmem_frontswap_flush_page, |
368 | .invalidate_area = tmem_frontswap_flush_area, | |
afec6e04 DM |
369 | .init = tmem_frontswap_init |
370 | }; | |
371 | #endif | |
372 | ||
373 | static int __init xen_tmem_init(void) | |
374 | { | |
5bc20fc5 DM |
375 | if (!xen_domain()) |
376 | return 0; | |
afec6e04 DM |
377 | #ifdef CONFIG_FRONTSWAP |
378 | if (tmem_enabled && use_frontswap) { | |
379 | char *s = ""; | |
380 | struct frontswap_ops old_ops = | |
381 | frontswap_register_ops(&tmem_frontswap_ops); | |
382 | ||
383 | tmem_frontswap_poolid = -1; | |
384 | if (old_ops.init != NULL) | |
385 | s = " (WARNING: frontswap_ops overridden)"; | |
386 | printk(KERN_INFO "frontswap enabled, RAM provided by " | |
387 | "Xen Transcendent Memory\n"); | |
388 | } | |
389 | #endif | |
5bc20fc5 DM |
390 | #ifdef CONFIG_CLEANCACHE |
391 | BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid)); | |
392 | if (tmem_enabled && use_cleancache) { | |
393 | char *s = ""; | |
afec6e04 DM |
394 | struct cleancache_ops old_ops = |
395 | cleancache_register_ops(&tmem_cleancache_ops); | |
5bc20fc5 DM |
396 | if (old_ops.init_fs != NULL) |
397 | s = " (WARNING: cleancache_ops overridden)"; | |
398 | printk(KERN_INFO "cleancache enabled, RAM provided by " | |
399 | "Xen Transcendent Memory%s\n", s); | |
400 | } | |
401 | #endif | |
402 | return 0; | |
403 | } | |
404 | ||
405 | module_init(xen_tmem_init) |