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 disable_cleancache __read_mostly;
36 static bool disable_selfballooning __read_mostly;
37 #ifdef CONFIG_XEN_TMEM_MODULE
38 module_param(disable_cleancache, bool, S_IRUGO);
39 module_param(disable_selfballooning, bool, S_IRUGO);
41 static int __init no_cleancache(char *s)
43 disable_cleancache = true;
46 __setup("nocleancache", no_cleancache);
48 #endif /* CONFIG_CLEANCACHE */
50 #ifdef CONFIG_FRONTSWAP
51 static bool disable_frontswap __read_mostly;
52 #ifdef CONFIG_XEN_TMEM_MODULE
53 module_param(disable_frontswap, bool, S_IRUGO);
55 static int __init no_frontswap(char *s)
57 disable_frontswap = true;
60 __setup("nofrontswap", no_frontswap);
62 #endif /* CONFIG_FRONTSWAP */
64 #ifdef CONFIG_FRONTSWAP
65 static bool disable_frontswap_selfshrinking __read_mostly;
66 #ifdef CONFIG_XEN_TMEM_MODULE
67 module_param(disable_frontswap_selfshrinking, bool, S_IRUGO);
69 #define disable_frontswap_selfshrinking 1
71 #endif /* CONFIG_FRONTSWAP */
73 #define TMEM_CONTROL 0
74 #define TMEM_NEW_POOL 1
75 #define TMEM_DESTROY_POOL 2
76 #define TMEM_NEW_PAGE 3
77 #define TMEM_PUT_PAGE 4
78 #define TMEM_GET_PAGE 5
79 #define TMEM_FLUSH_PAGE 6
80 #define TMEM_FLUSH_OBJECT 7
85 /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */
86 #define TMEM_POOL_PERSIST 1
87 #define TMEM_POOL_SHARED 2
88 #define TMEM_POOL_PAGESIZE_SHIFT 4
89 #define TMEM_VERSION_SHIFT 24
92 struct tmem_pool_uuid {
101 #define TMEM_POOL_PRIVATE_UUID { 0, 0 }
103 /* flags for tmem_ops.new_pool */
104 #define TMEM_POOL_PERSIST 1
105 #define TMEM_POOL_SHARED 2
107 /* xen tmem foundation ops/hypercalls */
109 static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid,
110 u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len)
116 op.pool_id = tmem_pool;
117 op.u.gen.oid[0] = oid.oid[0];
118 op.u.gen.oid[1] = oid.oid[1];
119 op.u.gen.oid[2] = oid.oid[2];
120 op.u.gen.index = index;
121 op.u.gen.tmem_offset = tmem_offset;
122 op.u.gen.pfn_offset = pfn_offset;
124 set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn);
125 rc = HYPERVISOR_tmem_op(&op);
129 static int xen_tmem_new_pool(struct tmem_pool_uuid uuid,
130 u32 flags, unsigned long pagesize)
133 int rc = 0, pageshift;
135 for (pageshift = 0; pagesize != 1; pageshift++)
137 flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT;
138 flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT;
139 op.cmd = TMEM_NEW_POOL;
140 op.u.new.uuid[0] = uuid.uuid_lo;
141 op.u.new.uuid[1] = uuid.uuid_hi;
142 op.u.new.flags = flags;
143 rc = HYPERVISOR_tmem_op(&op);
147 /* xen generic tmem ops */
149 static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid,
150 u32 index, unsigned long pfn)
152 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
154 return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index,
158 static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid,
159 u32 index, unsigned long pfn)
161 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
163 return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index,
167 static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index)
169 return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index,
173 static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid)
175 return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0);
179 #ifdef CONFIG_CLEANCACHE
180 static int xen_tmem_destroy_pool(u32 pool_id)
182 struct tmem_oid oid = { { 0 } };
184 return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0);
189 static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key,
190 pgoff_t index, struct page *page)
192 u32 ind = (u32) index;
193 struct tmem_oid oid = *(struct tmem_oid *)&key;
194 unsigned long pfn = page_to_pfn(page);
200 mb(); /* ensure page is quiescent; tmem may address it with an alias */
201 (void)xen_tmem_put_page((u32)pool, oid, ind, pfn);
204 static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key,
205 pgoff_t index, struct page *page)
207 u32 ind = (u32) index;
208 struct tmem_oid oid = *(struct tmem_oid *)&key;
209 unsigned long pfn = page_to_pfn(page);
212 /* translate return values to linux semantics */
217 ret = xen_tmem_get_page((u32)pool, oid, ind, pfn);
224 static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key,
227 u32 ind = (u32) index;
228 struct tmem_oid oid = *(struct tmem_oid *)&key;
234 (void)xen_tmem_flush_page((u32)pool, oid, ind);
237 static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key)
239 struct tmem_oid oid = *(struct tmem_oid *)&key;
243 (void)xen_tmem_flush_object((u32)pool, oid);
246 static void tmem_cleancache_flush_fs(int pool)
250 (void)xen_tmem_destroy_pool((u32)pool);
253 static int tmem_cleancache_init_fs(size_t pagesize)
255 struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID;
257 return xen_tmem_new_pool(uuid_private, 0, pagesize);
260 static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize)
262 struct tmem_pool_uuid shared_uuid;
264 shared_uuid.uuid_lo = *(u64 *)uuid;
265 shared_uuid.uuid_hi = *(u64 *)(&uuid[8]);
266 return xen_tmem_new_pool(shared_uuid, TMEM_POOL_SHARED, pagesize);
269 static struct cleancache_ops tmem_cleancache_ops = {
270 .put_page = tmem_cleancache_put_page,
271 .get_page = tmem_cleancache_get_page,
272 .invalidate_page = tmem_cleancache_flush_page,
273 .invalidate_inode = tmem_cleancache_flush_inode,
274 .invalidate_fs = tmem_cleancache_flush_fs,
275 .init_shared_fs = tmem_cleancache_init_shared_fs,
276 .init_fs = tmem_cleancache_init_fs
280 #ifdef CONFIG_FRONTSWAP
281 /* frontswap tmem operations */
283 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
284 static int tmem_frontswap_poolid;
287 * Swizzling increases objects per swaptype, increasing tmem concurrency
288 * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS
291 #define SWIZ_MASK ((1 << SWIZ_BITS) - 1)
292 #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
293 #define iswiz(_ind) (_ind >> SWIZ_BITS)
295 static inline struct tmem_oid oswiz(unsigned type, u32 ind)
297 struct tmem_oid oid = { .oid = { 0 } };
298 oid.oid[0] = _oswiz(type, ind);
302 /* returns 0 if the page was successfully put into frontswap, -1 if not */
303 static int tmem_frontswap_store(unsigned type, pgoff_t offset,
306 u64 ind64 = (u64)offset;
307 u32 ind = (u32)offset;
308 unsigned long pfn = page_to_pfn(page);
309 int pool = tmem_frontswap_poolid;
316 mb(); /* ensure page is quiescent; tmem may address it with an alias */
317 ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn);
318 /* translate Xen tmem return values to linux semantics */
326 * returns 0 if the page was successfully gotten from frontswap, -1 if
327 * was not present (should never happen!)
329 static int tmem_frontswap_load(unsigned type, pgoff_t offset,
332 u64 ind64 = (u64)offset;
333 u32 ind = (u32)offset;
334 unsigned long pfn = page_to_pfn(page);
335 int pool = tmem_frontswap_poolid;
342 ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn);
343 /* translate Xen tmem return values to linux semantics */
350 /* flush a single page from frontswap */
351 static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset)
353 u64 ind64 = (u64)offset;
354 u32 ind = (u32)offset;
355 int pool = tmem_frontswap_poolid;
361 (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind));
364 /* flush all pages from the passed swaptype */
365 static void tmem_frontswap_flush_area(unsigned type)
367 int pool = tmem_frontswap_poolid;
372 for (ind = SWIZ_MASK; ind >= 0; ind--)
373 (void)xen_tmem_flush_object(pool, oswiz(type, ind));
376 static void tmem_frontswap_init(unsigned ignored)
378 struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID;
380 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
381 if (tmem_frontswap_poolid < 0)
382 tmem_frontswap_poolid =
383 xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE);
386 static struct frontswap_ops tmem_frontswap_ops = {
387 .store = tmem_frontswap_store,
388 .load = tmem_frontswap_load,
389 .invalidate_page = tmem_frontswap_flush_page,
390 .invalidate_area = tmem_frontswap_flush_area,
391 .init = tmem_frontswap_init
395 static int xen_tmem_init(void)
399 #ifdef CONFIG_FRONTSWAP
400 if (tmem_enabled && !disable_frontswap) {
402 struct frontswap_ops *old_ops =
403 frontswap_register_ops(&tmem_frontswap_ops);
405 tmem_frontswap_poolid = -1;
406 if (IS_ERR(old_ops) || old_ops) {
408 return PTR_ERR(old_ops);
409 s = " (WARNING: frontswap_ops overridden)";
411 printk(KERN_INFO "frontswap enabled, RAM provided by "
412 "Xen Transcendent Memory%s\n", s);
415 #ifdef CONFIG_CLEANCACHE
416 BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid));
417 if (tmem_enabled && !disable_cleancache) {
419 struct cleancache_ops *old_ops =
420 cleancache_register_ops(&tmem_cleancache_ops);
422 s = " (WARNING: cleancache_ops overridden)";
423 printk(KERN_INFO "cleancache enabled, RAM provided by "
424 "Xen Transcendent Memory%s\n", s);
427 #ifdef CONFIG_XEN_SELFBALLOONING
428 xen_selfballoon_init(!disable_selfballooning,
429 !disable_frontswap_selfshrinking);
434 module_init(xen_tmem_init)
435 MODULE_LICENSE("GPL");
436 MODULE_AUTHOR("Dan Magenheimer <dan.magenheimer@oracle.com>");
437 MODULE_DESCRIPTION("Shim to Xen transcendent memory");