1 /*
2 * Xen implementation for transcendent memory (tmem)
3 *
4 * Copyright (C) 2009-2011 Oracle Corp. All rights reserved.
5 * Author: Dan Magenheimer
6 */
7
8 #include <linux/module.h>
9 #include <linux/kernel.h>
10 #include <linux/types.h>
11 #include <linux/init.h>
12 #include <linux/pagemap.h>
13 #include <linux/cleancache.h>
14 #include <linux/frontswap.h>
15
16 #include <xen/xen.h>
17 #include <xen/interface/xen.h>
18 #include <asm/xen/hypercall.h>
19 #include <asm/xen/page.h>
20 #include <asm/xen/hypervisor.h>
21 #include <xen/tmem.h>
22
23 #ifndef CONFIG_XEN_TMEM_MODULE
24 bool __read_mostly tmem_enabled = false;
25
enable_tmem(char * s)26 static int __init enable_tmem(char *s)
27 {
28 tmem_enabled = true;
29 return 1;
30 }
31 __setup("tmem", enable_tmem);
32 #endif
33
34 #ifdef CONFIG_CLEANCACHE
35 static bool cleancache __read_mostly = true;
36 module_param(cleancache, bool, S_IRUGO);
37 static bool selfballooning __read_mostly = true;
38 module_param(selfballooning, bool, S_IRUGO);
39 #endif /* CONFIG_CLEANCACHE */
40
41 #ifdef CONFIG_FRONTSWAP
42 static bool frontswap __read_mostly = true;
43 module_param(frontswap, bool, S_IRUGO);
44 #else /* CONFIG_FRONTSWAP */
45 #define frontswap (0)
46 #endif /* CONFIG_FRONTSWAP */
47
48 #ifdef CONFIG_XEN_SELFBALLOONING
49 static bool selfshrinking __read_mostly = true;
50 module_param(selfshrinking, bool, S_IRUGO);
51 #endif /* CONFIG_XEN_SELFBALLOONING */
52
53 #define TMEM_CONTROL 0
54 #define TMEM_NEW_POOL 1
55 #define TMEM_DESTROY_POOL 2
56 #define TMEM_NEW_PAGE 3
57 #define TMEM_PUT_PAGE 4
58 #define TMEM_GET_PAGE 5
59 #define TMEM_FLUSH_PAGE 6
60 #define TMEM_FLUSH_OBJECT 7
61 #define TMEM_READ 8
62 #define TMEM_WRITE 9
63 #define TMEM_XCHG 10
64
65 /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */
66 #define TMEM_POOL_PERSIST 1
67 #define TMEM_POOL_SHARED 2
68 #define TMEM_POOL_PAGESIZE_SHIFT 4
69 #define TMEM_VERSION_SHIFT 24
70
71
72 struct tmem_pool_uuid {
73 u64 uuid_lo;
74 u64 uuid_hi;
75 };
76
77 struct tmem_oid {
78 u64 oid[3];
79 };
80
81 #define TMEM_POOL_PRIVATE_UUID { 0, 0 }
82
83 /* flags for tmem_ops.new_pool */
84 #define TMEM_POOL_PERSIST 1
85 #define TMEM_POOL_SHARED 2
86
87 /* xen tmem foundation ops/hypercalls */
88
xen_tmem_op(u32 tmem_cmd,u32 tmem_pool,struct tmem_oid oid,u32 index,unsigned long gmfn,u32 tmem_offset,u32 pfn_offset,u32 len)89 static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid,
90 u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len)
91 {
92 struct tmem_op op;
93 int rc = 0;
94
95 op.cmd = tmem_cmd;
96 op.pool_id = tmem_pool;
97 op.u.gen.oid[0] = oid.oid[0];
98 op.u.gen.oid[1] = oid.oid[1];
99 op.u.gen.oid[2] = oid.oid[2];
100 op.u.gen.index = index;
101 op.u.gen.tmem_offset = tmem_offset;
102 op.u.gen.pfn_offset = pfn_offset;
103 op.u.gen.len = len;
104 set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn);
105 rc = HYPERVISOR_tmem_op(&op);
106 return rc;
107 }
108
xen_tmem_new_pool(struct tmem_pool_uuid uuid,u32 flags,unsigned long pagesize)109 static int xen_tmem_new_pool(struct tmem_pool_uuid uuid,
110 u32 flags, unsigned long pagesize)
111 {
112 struct tmem_op op;
113 int rc = 0, pageshift;
114
115 for (pageshift = 0; pagesize != 1; pageshift++)
116 pagesize >>= 1;
117 flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT;
118 flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT;
119 op.cmd = TMEM_NEW_POOL;
120 op.u.new.uuid[0] = uuid.uuid_lo;
121 op.u.new.uuid[1] = uuid.uuid_hi;
122 op.u.new.flags = flags;
123 rc = HYPERVISOR_tmem_op(&op);
124 return rc;
125 }
126
127 /* xen generic tmem ops */
128
xen_tmem_put_page(u32 pool_id,struct tmem_oid oid,u32 index,unsigned long pfn)129 static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid,
130 u32 index, unsigned long pfn)
131 {
132 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
133
134 return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index,
135 gmfn, 0, 0, 0);
136 }
137
xen_tmem_get_page(u32 pool_id,struct tmem_oid oid,u32 index,unsigned long pfn)138 static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid,
139 u32 index, unsigned long pfn)
140 {
141 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
142
143 return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index,
144 gmfn, 0, 0, 0);
145 }
146
xen_tmem_flush_page(u32 pool_id,struct tmem_oid oid,u32 index)147 static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index)
148 {
149 return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index,
150 0, 0, 0, 0);
151 }
152
xen_tmem_flush_object(u32 pool_id,struct tmem_oid oid)153 static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid)
154 {
155 return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0);
156 }
157
158
159 #ifdef CONFIG_CLEANCACHE
xen_tmem_destroy_pool(u32 pool_id)160 static int xen_tmem_destroy_pool(u32 pool_id)
161 {
162 struct tmem_oid oid = { { 0 } };
163
164 return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0);
165 }
166
167 /* cleancache ops */
168
tmem_cleancache_put_page(int pool,struct cleancache_filekey key,pgoff_t index,struct page * page)169 static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key,
170 pgoff_t index, struct page *page)
171 {
172 u32 ind = (u32) index;
173 struct tmem_oid oid = *(struct tmem_oid *)&key;
174 unsigned long pfn = page_to_pfn(page);
175
176 if (pool < 0)
177 return;
178 if (ind != index)
179 return;
180 mb(); /* ensure page is quiescent; tmem may address it with an alias */
181 (void)xen_tmem_put_page((u32)pool, oid, ind, pfn);
182 }
183
tmem_cleancache_get_page(int pool,struct cleancache_filekey key,pgoff_t index,struct page * page)184 static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key,
185 pgoff_t index, struct page *page)
186 {
187 u32 ind = (u32) index;
188 struct tmem_oid oid = *(struct tmem_oid *)&key;
189 unsigned long pfn = page_to_pfn(page);
190 int ret;
191
192 /* translate return values to linux semantics */
193 if (pool < 0)
194 return -1;
195 if (ind != index)
196 return -1;
197 ret = xen_tmem_get_page((u32)pool, oid, ind, pfn);
198 if (ret == 1)
199 return 0;
200 else
201 return -1;
202 }
203
tmem_cleancache_flush_page(int pool,struct cleancache_filekey key,pgoff_t index)204 static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key,
205 pgoff_t index)
206 {
207 u32 ind = (u32) index;
208 struct tmem_oid oid = *(struct tmem_oid *)&key;
209
210 if (pool < 0)
211 return;
212 if (ind != index)
213 return;
214 (void)xen_tmem_flush_page((u32)pool, oid, ind);
215 }
216
tmem_cleancache_flush_inode(int pool,struct cleancache_filekey key)217 static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key)
218 {
219 struct tmem_oid oid = *(struct tmem_oid *)&key;
220
221 if (pool < 0)
222 return;
223 (void)xen_tmem_flush_object((u32)pool, oid);
224 }
225
tmem_cleancache_flush_fs(int pool)226 static void tmem_cleancache_flush_fs(int pool)
227 {
228 if (pool < 0)
229 return;
230 (void)xen_tmem_destroy_pool((u32)pool);
231 }
232
tmem_cleancache_init_fs(size_t pagesize)233 static int tmem_cleancache_init_fs(size_t pagesize)
234 {
235 struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID;
236
237 return xen_tmem_new_pool(uuid_private, 0, pagesize);
238 }
239
tmem_cleancache_init_shared_fs(char * uuid,size_t pagesize)240 static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize)
241 {
242 struct tmem_pool_uuid shared_uuid;
243
244 shared_uuid.uuid_lo = *(u64 *)uuid;
245 shared_uuid.uuid_hi = *(u64 *)(&uuid[8]);
246 return xen_tmem_new_pool(shared_uuid, TMEM_POOL_SHARED, pagesize);
247 }
248
249 static struct cleancache_ops tmem_cleancache_ops = {
250 .put_page = tmem_cleancache_put_page,
251 .get_page = tmem_cleancache_get_page,
252 .invalidate_page = tmem_cleancache_flush_page,
253 .invalidate_inode = tmem_cleancache_flush_inode,
254 .invalidate_fs = tmem_cleancache_flush_fs,
255 .init_shared_fs = tmem_cleancache_init_shared_fs,
256 .init_fs = tmem_cleancache_init_fs
257 };
258 #endif
259
260 #ifdef CONFIG_FRONTSWAP
261 /* frontswap tmem operations */
262
263 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
264 static int tmem_frontswap_poolid;
265
266 /*
267 * Swizzling increases objects per swaptype, increasing tmem concurrency
268 * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS
269 */
270 #define SWIZ_BITS 4
271 #define SWIZ_MASK ((1 << SWIZ_BITS) - 1)
272 #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
273 #define iswiz(_ind) (_ind >> SWIZ_BITS)
274
oswiz(unsigned type,u32 ind)275 static inline struct tmem_oid oswiz(unsigned type, u32 ind)
276 {
277 struct tmem_oid oid = { .oid = { 0 } };
278 oid.oid[0] = _oswiz(type, ind);
279 return oid;
280 }
281
282 /* returns 0 if the page was successfully put into frontswap, -1 if not */
tmem_frontswap_store(unsigned type,pgoff_t offset,struct page * page)283 static int tmem_frontswap_store(unsigned type, pgoff_t offset,
284 struct page *page)
285 {
286 u64 ind64 = (u64)offset;
287 u32 ind = (u32)offset;
288 unsigned long pfn = page_to_pfn(page);
289 int pool = tmem_frontswap_poolid;
290 int ret;
291
292 if (pool < 0)
293 return -1;
294 if (ind64 != ind)
295 return -1;
296 mb(); /* ensure page is quiescent; tmem may address it with an alias */
297 ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn);
298 /* translate Xen tmem return values to linux semantics */
299 if (ret == 1)
300 return 0;
301 else
302 return -1;
303 }
304
305 /*
306 * returns 0 if the page was successfully gotten from frontswap, -1 if
307 * was not present (should never happen!)
308 */
tmem_frontswap_load(unsigned type,pgoff_t offset,struct page * page)309 static int tmem_frontswap_load(unsigned type, pgoff_t offset,
310 struct page *page)
311 {
312 u64 ind64 = (u64)offset;
313 u32 ind = (u32)offset;
314 unsigned long pfn = page_to_pfn(page);
315 int pool = tmem_frontswap_poolid;
316 int ret;
317
318 if (pool < 0)
319 return -1;
320 if (ind64 != ind)
321 return -1;
322 ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn);
323 /* translate Xen tmem return values to linux semantics */
324 if (ret == 1)
325 return 0;
326 else
327 return -1;
328 }
329
330 /* flush a single page from frontswap */
tmem_frontswap_flush_page(unsigned type,pgoff_t offset)331 static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset)
332 {
333 u64 ind64 = (u64)offset;
334 u32 ind = (u32)offset;
335 int pool = tmem_frontswap_poolid;
336
337 if (pool < 0)
338 return;
339 if (ind64 != ind)
340 return;
341 (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind));
342 }
343
344 /* flush all pages from the passed swaptype */
tmem_frontswap_flush_area(unsigned type)345 static void tmem_frontswap_flush_area(unsigned type)
346 {
347 int pool = tmem_frontswap_poolid;
348 int ind;
349
350 if (pool < 0)
351 return;
352 for (ind = SWIZ_MASK; ind >= 0; ind--)
353 (void)xen_tmem_flush_object(pool, oswiz(type, ind));
354 }
355
tmem_frontswap_init(unsigned ignored)356 static void tmem_frontswap_init(unsigned ignored)
357 {
358 struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID;
359
360 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
361 if (tmem_frontswap_poolid < 0)
362 tmem_frontswap_poolid =
363 xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE);
364 }
365
366 static struct frontswap_ops tmem_frontswap_ops = {
367 .store = tmem_frontswap_store,
368 .load = tmem_frontswap_load,
369 .invalidate_page = tmem_frontswap_flush_page,
370 .invalidate_area = tmem_frontswap_flush_area,
371 .init = tmem_frontswap_init
372 };
373 #endif
374
xen_tmem_init(void)375 static int xen_tmem_init(void)
376 {
377 if (!xen_domain())
378 return 0;
379 #ifdef CONFIG_FRONTSWAP
380 if (tmem_enabled && frontswap) {
381 char *s = "";
382 struct frontswap_ops *old_ops;
383
384 tmem_frontswap_poolid = -1;
385 old_ops = frontswap_register_ops(&tmem_frontswap_ops);
386 if (IS_ERR(old_ops) || old_ops) {
387 if (IS_ERR(old_ops))
388 return PTR_ERR(old_ops);
389 s = " (WARNING: frontswap_ops overridden)";
390 }
391 printk(KERN_INFO "frontswap enabled, RAM provided by "
392 "Xen Transcendent Memory%s\n", s);
393 }
394 #endif
395 #ifdef CONFIG_CLEANCACHE
396 BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid));
397 if (tmem_enabled && cleancache) {
398 char *s = "";
399 struct cleancache_ops *old_ops =
400 cleancache_register_ops(&tmem_cleancache_ops);
401 if (old_ops)
402 s = " (WARNING: cleancache_ops overridden)";
403 printk(KERN_INFO "cleancache enabled, RAM provided by "
404 "Xen Transcendent Memory%s\n", s);
405 }
406 #endif
407 #ifdef CONFIG_XEN_SELFBALLOONING
408 /*
409 * There is no point of driving pages to the swap system if they
410 * aren't going anywhere in tmem universe.
411 */
412 if (!frontswap) {
413 selfshrinking = false;
414 selfballooning = false;
415 }
416 xen_selfballoon_init(selfballooning, selfshrinking);
417 #endif
418 return 0;
419 }
420
421 module_init(xen_tmem_init)
422 MODULE_LICENSE("GPL");
423 MODULE_AUTHOR("Dan Magenheimer <dan.magenheimer@oracle.com>");
424 MODULE_DESCRIPTION("Shim to Xen transcendent memory");
425