2 * Xen implementation for transcendent memory (tmem)
4 * Copyright (C) 2009-2011 Oracle Corp. All rights reserved.
5 * Author: Dan Magenheimer
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>
17 #include <xen/interface/xen.h>
18 #include <asm/xen/hypercall.h>
19 #include <asm/xen/page.h>
20 #include <asm/xen/hypervisor.h>
23 #ifndef CONFIG_XEN_TMEM_MODULE
24 bool __read_mostly tmem_enabled = false;
26 static int __init enable_tmem(char *s)
31 __setup("tmem", enable_tmem);
34 #ifdef CONFIG_CLEANCACHE
35 static bool cleancache __read_mostly = true;
36 static bool selfballooning __read_mostly = true;
37 #ifdef CONFIG_XEN_TMEM_MODULE
38 module_param(cleancache, bool, S_IRUGO);
39 module_param(selfballooning, bool, S_IRUGO);
41 static int __init no_cleancache(char *s)
46 __setup("nocleancache", no_cleancache);
48 #endif /* CONFIG_CLEANCACHE */
50 #ifdef CONFIG_FRONTSWAP
51 static bool frontswap __read_mostly = true;
52 #ifdef CONFIG_XEN_TMEM_MODULE
53 module_param(frontswap, bool, S_IRUGO);
55 static int __init no_frontswap(char *s)
60 __setup("nofrontswap", no_frontswap);
62 #endif /* CONFIG_FRONTSWAP */
64 #ifdef CONFIG_XEN_SELFBALLOONING
65 static bool frontswap_selfshrinking __read_mostly = true;
66 #ifdef CONFIG_XEN_TMEM_MODULE
67 module_param(frontswap_selfshrinking, bool, S_IRUGO);
69 #endif /* CONFIG_XEN_SELFBALLOONING */
71 #define TMEM_CONTROL 0
72 #define TMEM_NEW_POOL 1
73 #define TMEM_DESTROY_POOL 2
74 #define TMEM_NEW_PAGE 3
75 #define TMEM_PUT_PAGE 4
76 #define TMEM_GET_PAGE 5
77 #define TMEM_FLUSH_PAGE 6
78 #define TMEM_FLUSH_OBJECT 7
83 /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */
84 #define TMEM_POOL_PERSIST 1
85 #define TMEM_POOL_SHARED 2
86 #define TMEM_POOL_PAGESIZE_SHIFT 4
87 #define TMEM_VERSION_SHIFT 24
90 struct tmem_pool_uuid {
99 #define TMEM_POOL_PRIVATE_UUID { 0, 0 }
101 /* flags for tmem_ops.new_pool */
102 #define TMEM_POOL_PERSIST 1
103 #define TMEM_POOL_SHARED 2
105 /* xen tmem foundation ops/hypercalls */
107 static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid,
108 u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len)
114 op.pool_id = tmem_pool;
115 op.u.gen.oid[0] = oid.oid[0];
116 op.u.gen.oid[1] = oid.oid[1];
117 op.u.gen.oid[2] = oid.oid[2];
118 op.u.gen.index = index;
119 op.u.gen.tmem_offset = tmem_offset;
120 op.u.gen.pfn_offset = pfn_offset;
122 set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn);
123 rc = HYPERVISOR_tmem_op(&op);
127 static int xen_tmem_new_pool(struct tmem_pool_uuid uuid,
128 u32 flags, unsigned long pagesize)
131 int rc = 0, pageshift;
133 for (pageshift = 0; pagesize != 1; pageshift++)
135 flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT;
136 flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT;
137 op.cmd = TMEM_NEW_POOL;
138 op.u.new.uuid[0] = uuid.uuid_lo;
139 op.u.new.uuid[1] = uuid.uuid_hi;
140 op.u.new.flags = flags;
141 rc = HYPERVISOR_tmem_op(&op);
145 /* xen generic tmem ops */
147 static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid,
148 u32 index, unsigned long pfn)
150 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
152 return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index,
156 static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid,
157 u32 index, unsigned long pfn)
159 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
161 return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index,
165 static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index)
167 return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index,
171 static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid)
173 return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0);
177 #ifdef CONFIG_CLEANCACHE
178 static int xen_tmem_destroy_pool(u32 pool_id)
180 struct tmem_oid oid = { { 0 } };
182 return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0);
187 static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key,
188 pgoff_t index, struct page *page)
190 u32 ind = (u32) index;
191 struct tmem_oid oid = *(struct tmem_oid *)&key;
192 unsigned long pfn = page_to_pfn(page);
198 mb(); /* ensure page is quiescent; tmem may address it with an alias */
199 (void)xen_tmem_put_page((u32)pool, oid, ind, pfn);
202 static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key,
203 pgoff_t index, struct page *page)
205 u32 ind = (u32) index;
206 struct tmem_oid oid = *(struct tmem_oid *)&key;
207 unsigned long pfn = page_to_pfn(page);
210 /* translate return values to linux semantics */
215 ret = xen_tmem_get_page((u32)pool, oid, ind, pfn);
222 static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key,
225 u32 ind = (u32) index;
226 struct tmem_oid oid = *(struct tmem_oid *)&key;
232 (void)xen_tmem_flush_page((u32)pool, oid, ind);
235 static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key)
237 struct tmem_oid oid = *(struct tmem_oid *)&key;
241 (void)xen_tmem_flush_object((u32)pool, oid);
244 static void tmem_cleancache_flush_fs(int pool)
248 (void)xen_tmem_destroy_pool((u32)pool);
251 static int tmem_cleancache_init_fs(size_t pagesize)
253 struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID;
255 return xen_tmem_new_pool(uuid_private, 0, pagesize);
258 static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize)
260 struct tmem_pool_uuid shared_uuid;
262 shared_uuid.uuid_lo = *(u64 *)uuid;
263 shared_uuid.uuid_hi = *(u64 *)(&uuid[8]);
264 return xen_tmem_new_pool(shared_uuid, TMEM_POOL_SHARED, pagesize);
267 static struct cleancache_ops tmem_cleancache_ops = {
268 .put_page = tmem_cleancache_put_page,
269 .get_page = tmem_cleancache_get_page,
270 .invalidate_page = tmem_cleancache_flush_page,
271 .invalidate_inode = tmem_cleancache_flush_inode,
272 .invalidate_fs = tmem_cleancache_flush_fs,
273 .init_shared_fs = tmem_cleancache_init_shared_fs,
274 .init_fs = tmem_cleancache_init_fs
278 #ifdef CONFIG_FRONTSWAP
279 /* frontswap tmem operations */
281 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
282 static int tmem_frontswap_poolid;
285 * Swizzling increases objects per swaptype, increasing tmem concurrency
286 * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS
289 #define SWIZ_MASK ((1 << SWIZ_BITS) - 1)
290 #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
291 #define iswiz(_ind) (_ind >> SWIZ_BITS)
293 static inline struct tmem_oid oswiz(unsigned type, u32 ind)
295 struct tmem_oid oid = { .oid = { 0 } };
296 oid.oid[0] = _oswiz(type, ind);
300 /* returns 0 if the page was successfully put into frontswap, -1 if not */
301 static int tmem_frontswap_store(unsigned type, pgoff_t offset,
304 u64 ind64 = (u64)offset;
305 u32 ind = (u32)offset;
306 unsigned long pfn = page_to_pfn(page);
307 int pool = tmem_frontswap_poolid;
314 mb(); /* ensure page is quiescent; tmem may address it with an alias */
315 ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn);
316 /* translate Xen tmem return values to linux semantics */
324 * returns 0 if the page was successfully gotten from frontswap, -1 if
325 * was not present (should never happen!)
327 static int tmem_frontswap_load(unsigned type, pgoff_t offset,
330 u64 ind64 = (u64)offset;
331 u32 ind = (u32)offset;
332 unsigned long pfn = page_to_pfn(page);
333 int pool = tmem_frontswap_poolid;
340 ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn);
341 /* translate Xen tmem return values to linux semantics */
348 /* flush a single page from frontswap */
349 static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset)
351 u64 ind64 = (u64)offset;
352 u32 ind = (u32)offset;
353 int pool = tmem_frontswap_poolid;
359 (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind));
362 /* flush all pages from the passed swaptype */
363 static void tmem_frontswap_flush_area(unsigned type)
365 int pool = tmem_frontswap_poolid;
370 for (ind = SWIZ_MASK; ind >= 0; ind--)
371 (void)xen_tmem_flush_object(pool, oswiz(type, ind));
374 static void tmem_frontswap_init(unsigned ignored)
376 struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID;
378 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
379 if (tmem_frontswap_poolid < 0)
380 tmem_frontswap_poolid =
381 xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE);
384 static struct frontswap_ops tmem_frontswap_ops = {
385 .store = tmem_frontswap_store,
386 .load = tmem_frontswap_load,
387 .invalidate_page = tmem_frontswap_flush_page,
388 .invalidate_area = tmem_frontswap_flush_area,
389 .init = tmem_frontswap_init
393 static int xen_tmem_init(void)
397 #ifdef CONFIG_FRONTSWAP
398 if (tmem_enabled && frontswap) {
400 struct frontswap_ops *old_ops =
401 frontswap_register_ops(&tmem_frontswap_ops);
403 tmem_frontswap_poolid = -1;
404 if (IS_ERR(old_ops) || old_ops) {
406 return PTR_ERR(old_ops);
407 s = " (WARNING: frontswap_ops overridden)";
409 printk(KERN_INFO "frontswap enabled, RAM provided by "
410 "Xen Transcendent Memory%s\n", s);
413 #ifdef CONFIG_CLEANCACHE
414 BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid));
415 if (tmem_enabled && cleancache) {
417 struct cleancache_ops *old_ops =
418 cleancache_register_ops(&tmem_cleancache_ops);
420 s = " (WARNING: cleancache_ops overridden)";
421 printk(KERN_INFO "cleancache enabled, RAM provided by "
422 "Xen Transcendent Memory%s\n", s);
425 #ifdef CONFIG_XEN_SELFBALLOONING
426 xen_selfballoon_init(selfballooning, frontswap_selfshrinking);
431 module_init(xen_tmem_init)
432 MODULE_LICENSE("GPL");
433 MODULE_AUTHOR("Dan Magenheimer <dan.magenheimer@oracle.com>");
434 MODULE_DESCRIPTION("Shim to Xen transcendent memory");