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> | |
b8b0f559 | 24 | #include <xen/tmem.h> |
5bc20fc5 DM |
25 | |
26 | #define TMEM_CONTROL 0 | |
27 | #define TMEM_NEW_POOL 1 | |
28 | #define TMEM_DESTROY_POOL 2 | |
29 | #define TMEM_NEW_PAGE 3 | |
30 | #define TMEM_PUT_PAGE 4 | |
31 | #define TMEM_GET_PAGE 5 | |
32 | #define TMEM_FLUSH_PAGE 6 | |
33 | #define TMEM_FLUSH_OBJECT 7 | |
34 | #define TMEM_READ 8 | |
35 | #define TMEM_WRITE 9 | |
36 | #define TMEM_XCHG 10 | |
37 | ||
38 | /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */ | |
39 | #define TMEM_POOL_PERSIST 1 | |
40 | #define TMEM_POOL_SHARED 2 | |
41 | #define TMEM_POOL_PAGESIZE_SHIFT 4 | |
42 | #define TMEM_VERSION_SHIFT 24 | |
43 | ||
44 | ||
45 | struct tmem_pool_uuid { | |
46 | u64 uuid_lo; | |
47 | u64 uuid_hi; | |
48 | }; | |
49 | ||
50 | struct tmem_oid { | |
51 | u64 oid[3]; | |
52 | }; | |
53 | ||
54 | #define TMEM_POOL_PRIVATE_UUID { 0, 0 } | |
55 | ||
56 | /* flags for tmem_ops.new_pool */ | |
57 | #define TMEM_POOL_PERSIST 1 | |
58 | #define TMEM_POOL_SHARED 2 | |
59 | ||
60 | /* xen tmem foundation ops/hypercalls */ | |
61 | ||
62 | static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid, | |
63 | u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len) | |
64 | { | |
65 | struct tmem_op op; | |
66 | int rc = 0; | |
67 | ||
68 | op.cmd = tmem_cmd; | |
69 | op.pool_id = tmem_pool; | |
70 | op.u.gen.oid[0] = oid.oid[0]; | |
71 | op.u.gen.oid[1] = oid.oid[1]; | |
72 | op.u.gen.oid[2] = oid.oid[2]; | |
73 | op.u.gen.index = index; | |
74 | op.u.gen.tmem_offset = tmem_offset; | |
75 | op.u.gen.pfn_offset = pfn_offset; | |
76 | op.u.gen.len = len; | |
77 | set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn); | |
78 | rc = HYPERVISOR_tmem_op(&op); | |
79 | return rc; | |
80 | } | |
81 | ||
82 | static int xen_tmem_new_pool(struct tmem_pool_uuid uuid, | |
83 | u32 flags, unsigned long pagesize) | |
84 | { | |
85 | struct tmem_op op; | |
86 | int rc = 0, pageshift; | |
87 | ||
88 | for (pageshift = 0; pagesize != 1; pageshift++) | |
89 | pagesize >>= 1; | |
90 | flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT; | |
91 | flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT; | |
92 | op.cmd = TMEM_NEW_POOL; | |
93 | op.u.new.uuid[0] = uuid.uuid_lo; | |
94 | op.u.new.uuid[1] = uuid.uuid_hi; | |
95 | op.u.new.flags = flags; | |
96 | rc = HYPERVISOR_tmem_op(&op); | |
97 | return rc; | |
98 | } | |
99 | ||
100 | /* xen generic tmem ops */ | |
101 | ||
102 | static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid, | |
103 | u32 index, unsigned long pfn) | |
104 | { | |
105 | unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn; | |
106 | ||
107 | return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index, | |
108 | gmfn, 0, 0, 0); | |
109 | } | |
110 | ||
111 | static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid, | |
112 | u32 index, unsigned long pfn) | |
113 | { | |
114 | unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn; | |
115 | ||
116 | return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index, | |
117 | gmfn, 0, 0, 0); | |
118 | } | |
119 | ||
120 | static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index) | |
121 | { | |
122 | return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index, | |
123 | 0, 0, 0, 0); | |
124 | } | |
125 | ||
126 | static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid) | |
127 | { | |
128 | return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0); | |
129 | } | |
130 | ||
8e6f7c23 | 131 | bool __read_mostly tmem_enabled = false; |
5bc20fc5 DM |
132 | |
133 | static int __init enable_tmem(char *s) | |
134 | { | |
8e6f7c23 | 135 | tmem_enabled = true; |
5bc20fc5 DM |
136 | return 1; |
137 | } | |
5bc20fc5 DM |
138 | __setup("tmem", enable_tmem); |
139 | ||
afec6e04 DM |
140 | #ifdef CONFIG_CLEANCACHE |
141 | static int xen_tmem_destroy_pool(u32 pool_id) | |
142 | { | |
143 | struct tmem_oid oid = { { 0 } }; | |
144 | ||
145 | return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0); | |
146 | } | |
147 | ||
5bc20fc5 DM |
148 | /* cleancache ops */ |
149 | ||
150 | static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key, | |
151 | pgoff_t index, struct page *page) | |
152 | { | |
153 | u32 ind = (u32) index; | |
154 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
155 | unsigned long pfn = page_to_pfn(page); | |
156 | ||
157 | if (pool < 0) | |
158 | return; | |
159 | if (ind != index) | |
160 | return; | |
161 | mb(); /* ensure page is quiescent; tmem may address it with an alias */ | |
162 | (void)xen_tmem_put_page((u32)pool, oid, ind, pfn); | |
163 | } | |
164 | ||
165 | static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key, | |
166 | pgoff_t index, struct page *page) | |
167 | { | |
168 | u32 ind = (u32) index; | |
169 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
170 | unsigned long pfn = page_to_pfn(page); | |
171 | int ret; | |
172 | ||
173 | /* translate return values to linux semantics */ | |
174 | if (pool < 0) | |
175 | return -1; | |
176 | if (ind != index) | |
177 | return -1; | |
178 | ret = xen_tmem_get_page((u32)pool, oid, ind, pfn); | |
179 | if (ret == 1) | |
180 | return 0; | |
181 | else | |
182 | return -1; | |
183 | } | |
184 | ||
185 | static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key, | |
186 | pgoff_t index) | |
187 | { | |
188 | u32 ind = (u32) index; | |
189 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
190 | ||
191 | if (pool < 0) | |
192 | return; | |
193 | if (ind != index) | |
194 | return; | |
195 | (void)xen_tmem_flush_page((u32)pool, oid, ind); | |
196 | } | |
197 | ||
198 | static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key) | |
199 | { | |
200 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
201 | ||
202 | if (pool < 0) | |
203 | return; | |
204 | (void)xen_tmem_flush_object((u32)pool, oid); | |
205 | } | |
206 | ||
207 | static void tmem_cleancache_flush_fs(int pool) | |
208 | { | |
209 | if (pool < 0) | |
210 | return; | |
211 | (void)xen_tmem_destroy_pool((u32)pool); | |
212 | } | |
213 | ||
214 | static int tmem_cleancache_init_fs(size_t pagesize) | |
215 | { | |
216 | struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID; | |
217 | ||
218 | return xen_tmem_new_pool(uuid_private, 0, pagesize); | |
219 | } | |
220 | ||
221 | static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize) | |
222 | { | |
223 | struct tmem_pool_uuid shared_uuid; | |
224 | ||
225 | shared_uuid.uuid_lo = *(u64 *)uuid; | |
226 | shared_uuid.uuid_hi = *(u64 *)(&uuid[8]); | |
227 | return xen_tmem_new_pool(shared_uuid, TMEM_POOL_SHARED, pagesize); | |
228 | } | |
229 | ||
8e6f7c23 | 230 | static bool __initdata use_cleancache = true; |
5bc20fc5 DM |
231 | |
232 | static int __init no_cleancache(char *s) | |
233 | { | |
8e6f7c23 | 234 | use_cleancache = false; |
5bc20fc5 DM |
235 | return 1; |
236 | } | |
5bc20fc5 DM |
237 | __setup("nocleancache", no_cleancache); |
238 | ||
8e6f7c23 | 239 | static struct cleancache_ops __initdata tmem_cleancache_ops = { |
5bc20fc5 DM |
240 | .put_page = tmem_cleancache_put_page, |
241 | .get_page = tmem_cleancache_get_page, | |
91c6cc9b DM |
242 | .invalidate_page = tmem_cleancache_flush_page, |
243 | .invalidate_inode = tmem_cleancache_flush_inode, | |
244 | .invalidate_fs = tmem_cleancache_flush_fs, | |
5bc20fc5 DM |
245 | .init_shared_fs = tmem_cleancache_init_shared_fs, |
246 | .init_fs = tmem_cleancache_init_fs | |
247 | }; | |
afec6e04 | 248 | #endif |
5bc20fc5 | 249 | |
afec6e04 DM |
250 | #ifdef CONFIG_FRONTSWAP |
251 | /* frontswap tmem operations */ | |
252 | ||
253 | /* a single tmem poolid is used for all frontswap "types" (swapfiles) */ | |
254 | static int tmem_frontswap_poolid; | |
255 | ||
256 | /* | |
257 | * Swizzling increases objects per swaptype, increasing tmem concurrency | |
258 | * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS | |
259 | */ | |
260 | #define SWIZ_BITS 4 | |
261 | #define SWIZ_MASK ((1 << SWIZ_BITS) - 1) | |
262 | #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK)) | |
263 | #define iswiz(_ind) (_ind >> SWIZ_BITS) | |
264 | ||
265 | static inline struct tmem_oid oswiz(unsigned type, u32 ind) | |
5bc20fc5 | 266 | { |
afec6e04 DM |
267 | struct tmem_oid oid = { .oid = { 0 } }; |
268 | oid.oid[0] = _oswiz(type, ind); | |
269 | return oid; | |
270 | } | |
5bc20fc5 | 271 | |
afec6e04 | 272 | /* returns 0 if the page was successfully put into frontswap, -1 if not */ |
165c8aed | 273 | static int tmem_frontswap_store(unsigned type, pgoff_t offset, |
afec6e04 DM |
274 | struct page *page) |
275 | { | |
276 | u64 ind64 = (u64)offset; | |
277 | u32 ind = (u32)offset; | |
278 | unsigned long pfn = page_to_pfn(page); | |
279 | int pool = tmem_frontswap_poolid; | |
280 | int ret; | |
281 | ||
282 | if (pool < 0) | |
283 | return -1; | |
284 | if (ind64 != ind) | |
285 | return -1; | |
286 | mb(); /* ensure page is quiescent; tmem may address it with an alias */ | |
287 | ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn); | |
288 | /* translate Xen tmem return values to linux semantics */ | |
289 | if (ret == 1) | |
290 | return 0; | |
291 | else | |
292 | return -1; | |
293 | } | |
294 | ||
295 | /* | |
296 | * returns 0 if the page was successfully gotten from frontswap, -1 if | |
297 | * was not present (should never happen!) | |
298 | */ | |
165c8aed | 299 | static int tmem_frontswap_load(unsigned type, pgoff_t offset, |
afec6e04 DM |
300 | struct page *page) |
301 | { | |
302 | u64 ind64 = (u64)offset; | |
303 | u32 ind = (u32)offset; | |
304 | unsigned long pfn = page_to_pfn(page); | |
305 | int pool = tmem_frontswap_poolid; | |
306 | int ret; | |
307 | ||
308 | if (pool < 0) | |
309 | return -1; | |
310 | if (ind64 != ind) | |
311 | return -1; | |
312 | ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn); | |
313 | /* translate Xen tmem return values to linux semantics */ | |
314 | if (ret == 1) | |
315 | return 0; | |
316 | else | |
317 | return -1; | |
318 | } | |
319 | ||
320 | /* flush a single page from frontswap */ | |
321 | static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset) | |
322 | { | |
323 | u64 ind64 = (u64)offset; | |
324 | u32 ind = (u32)offset; | |
325 | int pool = tmem_frontswap_poolid; | |
326 | ||
327 | if (pool < 0) | |
328 | return; | |
329 | if (ind64 != ind) | |
330 | return; | |
331 | (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind)); | |
332 | } | |
333 | ||
334 | /* flush all pages from the passed swaptype */ | |
335 | static void tmem_frontswap_flush_area(unsigned type) | |
336 | { | |
337 | int pool = tmem_frontswap_poolid; | |
338 | int ind; | |
339 | ||
340 | if (pool < 0) | |
341 | return; | |
342 | for (ind = SWIZ_MASK; ind >= 0; ind--) | |
343 | (void)xen_tmem_flush_object(pool, oswiz(type, ind)); | |
344 | } | |
345 | ||
346 | static void tmem_frontswap_init(unsigned ignored) | |
347 | { | |
348 | struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID; | |
349 | ||
350 | /* a single tmem poolid is used for all frontswap "types" (swapfiles) */ | |
351 | if (tmem_frontswap_poolid < 0) | |
352 | tmem_frontswap_poolid = | |
353 | xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE); | |
354 | } | |
355 | ||
8e6f7c23 | 356 | static bool __initdata use_frontswap = true; |
afec6e04 DM |
357 | |
358 | static int __init no_frontswap(char *s) | |
359 | { | |
8e6f7c23 | 360 | use_frontswap = false; |
afec6e04 DM |
361 | return 1; |
362 | } | |
afec6e04 DM |
363 | __setup("nofrontswap", no_frontswap); |
364 | ||
8e6f7c23 | 365 | static struct frontswap_ops __initdata tmem_frontswap_ops = { |
165c8aed KRW |
366 | .store = tmem_frontswap_store, |
367 | .load = tmem_frontswap_load, | |
91c6cc9b DM |
368 | .invalidate_page = tmem_frontswap_flush_page, |
369 | .invalidate_area = tmem_frontswap_flush_area, | |
afec6e04 DM |
370 | .init = tmem_frontswap_init |
371 | }; | |
372 | #endif | |
373 | ||
374 | static int __init xen_tmem_init(void) | |
375 | { | |
5bc20fc5 DM |
376 | if (!xen_domain()) |
377 | return 0; | |
afec6e04 DM |
378 | #ifdef CONFIG_FRONTSWAP |
379 | if (tmem_enabled && use_frontswap) { | |
380 | char *s = ""; | |
381 | struct frontswap_ops old_ops = | |
382 | frontswap_register_ops(&tmem_frontswap_ops); | |
383 | ||
384 | tmem_frontswap_poolid = -1; | |
385 | if (old_ops.init != NULL) | |
386 | s = " (WARNING: frontswap_ops overridden)"; | |
387 | printk(KERN_INFO "frontswap enabled, RAM provided by " | |
22230c18 | 388 | "Xen Transcendent Memory%s\n", s); |
afec6e04 DM |
389 | } |
390 | #endif | |
5bc20fc5 DM |
391 | #ifdef CONFIG_CLEANCACHE |
392 | BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid)); | |
393 | if (tmem_enabled && use_cleancache) { | |
394 | char *s = ""; | |
afec6e04 DM |
395 | struct cleancache_ops old_ops = |
396 | cleancache_register_ops(&tmem_cleancache_ops); | |
5bc20fc5 DM |
397 | if (old_ops.init_fs != NULL) |
398 | s = " (WARNING: cleancache_ops overridden)"; | |
399 | printk(KERN_INFO "cleancache enabled, RAM provided by " | |
400 | "Xen Transcendent Memory%s\n", s); | |
401 | } | |
402 | #endif | |
403 | return 0; | |
404 | } | |
405 | ||
406 | module_init(xen_tmem_init) |