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 #define TMEM_CONTROL 0
24 #define TMEM_NEW_POOL 1
25 #define TMEM_DESTROY_POOL 2
26 #define TMEM_NEW_PAGE 3
27 #define TMEM_PUT_PAGE 4
28 #define TMEM_GET_PAGE 5
29 #define TMEM_FLUSH_PAGE 6
30 #define TMEM_FLUSH_OBJECT 7
35 /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */
36 #define TMEM_POOL_PERSIST 1
37 #define TMEM_POOL_SHARED 2
38 #define TMEM_POOL_PAGESIZE_SHIFT 4
39 #define TMEM_VERSION_SHIFT 24
42 struct tmem_pool_uuid {
51 #define TMEM_POOL_PRIVATE_UUID { 0, 0 }
53 /* flags for tmem_ops.new_pool */
54 #define TMEM_POOL_PERSIST 1
55 #define TMEM_POOL_SHARED 2
57 /* xen tmem foundation ops/hypercalls */
59 static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid,
60 u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len)
66 op.pool_id = tmem_pool;
67 op.u.gen.oid[0] = oid.oid[0];
68 op.u.gen.oid[1] = oid.oid[1];
69 op.u.gen.oid[2] = oid.oid[2];
70 op.u.gen.index = index;
71 op.u.gen.tmem_offset = tmem_offset;
72 op.u.gen.pfn_offset = pfn_offset;
74 set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn);
75 rc = HYPERVISOR_tmem_op(&op);
79 static int xen_tmem_new_pool(struct tmem_pool_uuid uuid,
80 u32 flags, unsigned long pagesize)
83 int rc = 0, pageshift;
85 for (pageshift = 0; pagesize != 1; pageshift++)
87 flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT;
88 flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT;
89 op.cmd = TMEM_NEW_POOL;
90 op.u.new.uuid[0] = uuid.uuid_lo;
91 op.u.new.uuid[1] = uuid.uuid_hi;
92 op.u.new.flags = flags;
93 rc = HYPERVISOR_tmem_op(&op);
97 /* xen generic tmem ops */
99 static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid,
100 u32 index, unsigned long pfn)
102 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
104 return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index,
108 static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid,
109 u32 index, unsigned long pfn)
111 unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
113 return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index,
117 static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index)
119 return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index,
123 static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid)
125 return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0);
128 #ifndef CONFIG_XEN_TMEM_MODULE
129 bool __read_mostly tmem_enabled = false;
131 static int __init enable_tmem(char *s)
136 __setup("tmem", enable_tmem);
139 #ifdef CONFIG_CLEANCACHE
140 static int xen_tmem_destroy_pool(u32 pool_id)
142 struct tmem_oid oid = { { 0 } };
144 return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0);
149 static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key,
150 pgoff_t index, struct page *page)
152 u32 ind = (u32) index;
153 struct tmem_oid oid = *(struct tmem_oid *)&key;
154 unsigned long pfn = page_to_pfn(page);
160 mb(); /* ensure page is quiescent; tmem may address it with an alias */
161 (void)xen_tmem_put_page((u32)pool, oid, ind, pfn);
164 static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key,
165 pgoff_t index, struct page *page)
167 u32 ind = (u32) index;
168 struct tmem_oid oid = *(struct tmem_oid *)&key;
169 unsigned long pfn = page_to_pfn(page);
172 /* translate return values to linux semantics */
177 ret = xen_tmem_get_page((u32)pool, oid, ind, pfn);
184 static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key,
187 u32 ind = (u32) index;
188 struct tmem_oid oid = *(struct tmem_oid *)&key;
194 (void)xen_tmem_flush_page((u32)pool, oid, ind);
197 static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key)
199 struct tmem_oid oid = *(struct tmem_oid *)&key;
203 (void)xen_tmem_flush_object((u32)pool, oid);
206 static void tmem_cleancache_flush_fs(int pool)
210 (void)xen_tmem_destroy_pool((u32)pool);
213 static int tmem_cleancache_init_fs(size_t pagesize)
215 struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID;
217 return xen_tmem_new_pool(uuid_private, 0, pagesize);
220 static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize)
222 struct tmem_pool_uuid shared_uuid;
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);
229 static bool disable_cleancache __read_mostly;
230 static bool disable_selfballooning __read_mostly;
231 #ifdef CONFIG_XEN_TMEM_MODULE
232 module_param(disable_cleancache, bool, S_IRUGO);
233 module_param(disable_selfballooning, bool, S_IRUGO);
235 static int __init no_cleancache(char *s)
237 disable_cleancache = true;
240 __setup("nocleancache", no_cleancache);
243 static struct cleancache_ops tmem_cleancache_ops = {
244 .put_page = tmem_cleancache_put_page,
245 .get_page = tmem_cleancache_get_page,
246 .invalidate_page = tmem_cleancache_flush_page,
247 .invalidate_inode = tmem_cleancache_flush_inode,
248 .invalidate_fs = tmem_cleancache_flush_fs,
249 .init_shared_fs = tmem_cleancache_init_shared_fs,
250 .init_fs = tmem_cleancache_init_fs
254 #ifdef CONFIG_FRONTSWAP
255 /* frontswap tmem operations */
257 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
258 static int tmem_frontswap_poolid;
261 * Swizzling increases objects per swaptype, increasing tmem concurrency
262 * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS
265 #define SWIZ_MASK ((1 << SWIZ_BITS) - 1)
266 #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
267 #define iswiz(_ind) (_ind >> SWIZ_BITS)
269 static inline struct tmem_oid oswiz(unsigned type, u32 ind)
271 struct tmem_oid oid = { .oid = { 0 } };
272 oid.oid[0] = _oswiz(type, ind);
276 /* returns 0 if the page was successfully put into frontswap, -1 if not */
277 static int tmem_frontswap_store(unsigned type, pgoff_t offset,
280 u64 ind64 = (u64)offset;
281 u32 ind = (u32)offset;
282 unsigned long pfn = page_to_pfn(page);
283 int pool = tmem_frontswap_poolid;
290 mb(); /* ensure page is quiescent; tmem may address it with an alias */
291 ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn);
292 /* translate Xen tmem return values to linux semantics */
300 * returns 0 if the page was successfully gotten from frontswap, -1 if
301 * was not present (should never happen!)
303 static int tmem_frontswap_load(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 ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn);
317 /* translate Xen tmem return values to linux semantics */
324 /* flush a single page from frontswap */
325 static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset)
327 u64 ind64 = (u64)offset;
328 u32 ind = (u32)offset;
329 int pool = tmem_frontswap_poolid;
335 (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind));
338 /* flush all pages from the passed swaptype */
339 static void tmem_frontswap_flush_area(unsigned type)
341 int pool = tmem_frontswap_poolid;
346 for (ind = SWIZ_MASK; ind >= 0; ind--)
347 (void)xen_tmem_flush_object(pool, oswiz(type, ind));
350 static void tmem_frontswap_init(unsigned ignored)
352 struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID;
354 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
355 if (tmem_frontswap_poolid < 0)
356 tmem_frontswap_poolid =
357 xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE);
360 static bool disable_frontswap __read_mostly;
361 static bool disable_frontswap_selfshrinking __read_mostly;
362 #ifdef CONFIG_XEN_TMEM_MODULE
363 module_param(disable_frontswap, bool, S_IRUGO);
364 module_param(disable_frontswap_selfshrinking, bool, S_IRUGO);
366 static int __init no_frontswap(char *s)
368 disable_frontswap = true;
371 __setup("nofrontswap", no_frontswap);
374 static struct frontswap_ops tmem_frontswap_ops = {
375 .store = tmem_frontswap_store,
376 .load = tmem_frontswap_load,
377 .invalidate_page = tmem_frontswap_flush_page,
378 .invalidate_area = tmem_frontswap_flush_area,
379 .init = tmem_frontswap_init
381 #else /* CONFIG_FRONTSWAP */
382 #define disable_frontswap_selfshrinking 1
385 static int xen_tmem_init(void)
389 #ifdef CONFIG_FRONTSWAP
390 if (tmem_enabled && !disable_frontswap) {
392 struct frontswap_ops *old_ops =
393 frontswap_register_ops(&tmem_frontswap_ops);
395 tmem_frontswap_poolid = -1;
396 if (IS_ERR(old_ops) || old_ops) {
398 return PTR_ERR(old_ops);
399 s = " (WARNING: frontswap_ops overridden)";
401 printk(KERN_INFO "frontswap enabled, RAM provided by "
402 "Xen Transcendent Memory%s\n", s);
405 #ifdef CONFIG_CLEANCACHE
406 BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid));
407 if (tmem_enabled && !disable_cleancache) {
409 struct cleancache_ops *old_ops =
410 cleancache_register_ops(&tmem_cleancache_ops);
412 s = " (WARNING: cleancache_ops overridden)";
413 printk(KERN_INFO "cleancache enabled, RAM provided by "
414 "Xen Transcendent Memory%s\n", s);
417 #ifdef CONFIG_XEN_SELFBALLOONING
418 xen_selfballoon_init(!disable_selfballooning,
419 !disable_frontswap_selfshrinking);
424 module_init(xen_tmem_init)
425 MODULE_LICENSE("GPL");
426 MODULE_AUTHOR("Dan Magenheimer <dan.magenheimer@oracle.com>");
427 MODULE_DESCRIPTION("Shim to Xen transcendent memory");