Bug Summary

File:kern/subr_hibernate.c
Warning:line 1535, column 14
The left operand of '==' is a garbage value

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -cc1 -triple amd64-unknown-openbsd7.0 -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name subr_hibernate.c -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -mrelocation-model static -mframe-pointer=all -relaxed-aliasing -fno-rounding-math -mconstructor-aliases -ffreestanding -mcmodel=kernel -target-cpu x86-64 -target-feature +retpoline-indirect-calls -target-feature +retpoline-indirect-branches -target-feature -sse2 -target-feature -sse -target-feature -3dnow -target-feature -mmx -target-feature +save-args -disable-red-zone -no-implicit-float -tune-cpu generic -debugger-tuning=gdb -fcoverage-compilation-dir=/usr/src/sys/arch/amd64/compile/GENERIC.MP/obj -nostdsysteminc -nobuiltininc -resource-dir /usr/local/lib/clang/13.0.0 -I /usr/src/sys -I /usr/src/sys/arch/amd64/compile/GENERIC.MP/obj -I /usr/src/sys/arch -I /usr/src/sys/dev/pci/drm/include -I /usr/src/sys/dev/pci/drm/include/uapi -I /usr/src/sys/dev/pci/drm/amd/include/asic_reg -I /usr/src/sys/dev/pci/drm/amd/include -I /usr/src/sys/dev/pci/drm/amd/amdgpu -I /usr/src/sys/dev/pci/drm/amd/display -I /usr/src/sys/dev/pci/drm/amd/display/include -I /usr/src/sys/dev/pci/drm/amd/display/dc -I /usr/src/sys/dev/pci/drm/amd/display/amdgpu_dm -I /usr/src/sys/dev/pci/drm/amd/pm/inc -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/smu11 -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/smu12 -I /usr/src/sys/dev/pci/drm/amd/pm/powerplay -I /usr/src/sys/dev/pci/drm/amd/pm/powerplay/hwmgr -I /usr/src/sys/dev/pci/drm/amd/pm/powerplay/smumgr -I /usr/src/sys/dev/pci/drm/amd/display/dc/inc -I /usr/src/sys/dev/pci/drm/amd/display/dc/inc/hw -I /usr/src/sys/dev/pci/drm/amd/display/dc/clk_mgr -I /usr/src/sys/dev/pci/drm/amd/display/modules/inc -I /usr/src/sys/dev/pci/drm/amd/display/modules/hdcp -I /usr/src/sys/dev/pci/drm/amd/display/dmub/inc -I /usr/src/sys/dev/pci/drm/i915 -D DDB -D DIAGNOSTIC -D KTRACE -D ACCOUNTING -D KMEMSTATS -D PTRACE -D POOL_DEBUG -D CRYPTO -D SYSVMSG -D SYSVSEM -D SYSVSHM -D UVM_SWAP_ENCRYPT -D FFS -D FFS2 -D FFS_SOFTUPDATES -D UFS_DIRHASH -D QUOTA -D EXT2FS -D MFS -D NFSCLIENT -D NFSSERVER -D CD9660 -D UDF -D MSDOSFS -D FIFO -D FUSE -D SOCKET_SPLICE -D TCP_ECN -D TCP_SIGNATURE -D INET6 -D IPSEC -D PPP_BSDCOMP -D PPP_DEFLATE -D PIPEX -D MROUTING -D MPLS -D BOOT_CONFIG -D USER_PCICONF -D APERTURE -D MTRR -D NTFS -D HIBERNATE -D PCIVERBOSE -D USBVERBOSE -D WSDISPLAY_COMPAT_USL -D WSDISPLAY_COMPAT_RAWKBD -D WSDISPLAY_DEFAULTSCREENS=6 -D X86EMU -D ONEWIREVERBOSE -D MULTIPROCESSOR -D MAXUSERS=80 -D _KERNEL -D CONFIG_DRM_AMD_DC_DCN3_0 -O2 -Wno-pointer-sign -Wno-address-of-packed-member -Wno-constant-conversion -Wno-unused-but-set-variable -Wno-gnu-folding-constant -fdebug-compilation-dir=/usr/src/sys/arch/amd64/compile/GENERIC.MP/obj -ferror-limit 19 -fwrapv -D_RET_PROTECTOR -ret-protector -fgnuc-version=4.2.1 -vectorize-loops -vectorize-slp -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-valloc -fno-builtin-free -fno-builtin-strdup -fno-builtin-strndup -analyzer-output=html -faddrsig -o /usr/obj/sys/arch/amd64/compile/GENERIC.MP/scan-build/2022-01-12-131800-47421-1 -x c /usr/src/sys/kern/subr_hibernate.c
1/* $OpenBSD: subr_hibernate.c,v 1.132 2022/01/07 02:47:07 guenther Exp $ */
2
3/*
4 * Copyright (c) 2011 Ariane van der Steldt <ariane@stack.nl>
5 * Copyright (c) 2011 Mike Larkin <mlarkin@openbsd.org>
6 *
7 * Permission to use, copy, modify, and distribute this software for any
8 * purpose with or without fee is hereby granted, provided that the above
9 * copyright notice and this permission notice appear in all copies.
10 *
11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18 */
19
20#include <sys/hibernate.h>
21#include <sys/malloc.h>
22#include <sys/param.h>
23#include <sys/tree.h>
24#include <sys/systm.h>
25#include <sys/disklabel.h>
26#include <sys/disk.h>
27#include <sys/conf.h>
28#include <sys/buf.h>
29#include <sys/fcntl.h>
30#include <sys/stat.h>
31#include <sys/atomic.h>
32
33#include <uvm/uvm.h>
34#include <uvm/uvm_swap.h>
35
36#include <machine/hibernate.h>
37
38/* Make sure the signature can fit in one block */
39CTASSERT(sizeof(union hibernate_info) <= DEV_BSIZE)extern char _ctassert[(sizeof(union hibernate_info) <= (1 <<
9)) ? 1 : -1 ] __attribute__((__unused__))
;
40
41/*
42 * Hibernate piglet layout information
43 *
44 * The piglet is a scratch area of memory allocated by the suspending kernel.
45 * Its phys and virt addrs are recorded in the signature block. The piglet is
46 * used to guarantee an unused area of memory that can be used by the resuming
47 * kernel for various things. The piglet is excluded during unpack operations.
48 * The piglet size is presently 4*HIBERNATE_CHUNK_SIZE (typically 4*4MB).
49 *
50 * Offset from piglet_base Purpose
51 * ----------------------------------------------------------------------------
52 * 0 Private page for suspend I/O write functions
53 * 1*PAGE_SIZE I/O page used during hibernate suspend
54 * 2*PAGE_SIZE I/O page used during hibernate suspend
55 * 3*PAGE_SIZE copy page used during hibernate suspend
56 * 4*PAGE_SIZE final chunk ordering list (24 pages)
57 * 28*PAGE_SIZE RLE utility page
58 * 29*PAGE_SIZE start of hiballoc area
59 * 30*PAGE_SIZE preserved entropy
60 * 110*PAGE_SIZE end of hiballoc area (80 pages)
61 * 366*PAGE_SIZE end of retguard preservation region (256 pages)
62 * ... unused
63 * HIBERNATE_CHUNK_SIZE start of hibernate chunk table
64 * 2*HIBERNATE_CHUNK_SIZE bounce area for chunks being unpacked
65 * 4*HIBERNATE_CHUNK_SIZE end of piglet
66 */
67
68/* Temporary vaddr ranges used during hibernate */
69vaddr_t hibernate_temp_page;
70vaddr_t hibernate_copy_page;
71vaddr_t hibernate_rle_page;
72
73/* Hibernate info as read from disk during resume */
74union hibernate_info disk_hib;
75
76/*
77 * Global copy of the pig start address. This needs to be a global as we
78 * switch stacks after computing it - it can't be stored on the stack.
79 */
80paddr_t global_pig_start;
81
82/*
83 * Global copies of the piglet start addresses (PA/VA). We store these
84 * as globals to avoid having to carry them around as parameters, as the
85 * piglet is allocated early and freed late - its lifecycle extends beyond
86 * that of the hibernate info union which is calculated on suspend/resume.
87 */
88vaddr_t global_piglet_va;
89paddr_t global_piglet_pa;
90
91/* #define HIB_DEBUG */
92#ifdef HIB_DEBUG
93int hib_debug = 99;
94#define DPRINTF(x...) do { if (hib_debug) printf(x); } while (0)
95#define DNPRINTF(n,x...) do { if (hib_debug > (n)) printf(x); } while (0)
96#else
97#define DPRINTF(x...)
98#define DNPRINTF(n,x...)
99#endif
100
101#ifndef NO_PROPOLICE
102extern long __guard_local;
103#endif /* ! NO_PROPOLICE */
104
105void hibernate_copy_chunk_to_piglet(paddr_t, vaddr_t, size_t);
106int hibernate_calc_rle(paddr_t, paddr_t);
107int hibernate_write_rle(union hibernate_info *, paddr_t, paddr_t, daddr_t *,
108 size_t *);
109
110#define MAX_RLE(0x400000 / (1 << 12)) (HIBERNATE_CHUNK_SIZE0x400000 / PAGE_SIZE(1 << 12))
111
112/*
113 * Hib alloc enforced alignment.
114 */
115#define HIB_ALIGN8 8 /* bytes alignment */
116
117/*
118 * sizeof builtin operation, but with alignment constraint.
119 */
120#define HIB_SIZEOF(_type)((((sizeof(_type))+((8)-1))/(8))*(8)) roundup(sizeof(_type), HIB_ALIGN)((((sizeof(_type))+((8)-1))/(8))*(8))
121
122struct hiballoc_entry {
123 size_t hibe_use;
124 size_t hibe_space;
125 RBT_ENTRY(hiballoc_entry)struct rb_entry hibe_entry;
126};
127
128/*
129 * Sort hibernate memory ranges by ascending PA
130 */
131void
132hibernate_sort_ranges(union hibernate_info *hib_info)
133{
134 int i, j;
135 struct hibernate_memory_range *ranges;
136 paddr_t base, end;
137
138 ranges = hib_info->ranges;
139
140 for (i = 1; i < hib_info->nranges; i++) {
141 j = i;
142 while (j > 0 && ranges[j - 1].base > ranges[j].base) {
143 base = ranges[j].base;
144 end = ranges[j].end;
145 ranges[j].base = ranges[j - 1].base;
146 ranges[j].end = ranges[j - 1].end;
147 ranges[j - 1].base = base;
148 ranges[j - 1].end = end;
149 j--;
150 }
151 }
152}
153
154/*
155 * Compare hiballoc entries based on the address they manage.
156 *
157 * Since the address is fixed, relative to struct hiballoc_entry,
158 * we just compare the hiballoc_entry pointers.
159 */
160static __inline int
161hibe_cmp(const struct hiballoc_entry *l, const struct hiballoc_entry *r)
162{
163 vaddr_t vl = (vaddr_t)l;
164 vaddr_t vr = (vaddr_t)r;
165
166 return vl < vr ? -1 : (vl > vr);
167}
168
169RBT_PROTOTYPE(hiballoc_addr, hiballoc_entry, hibe_entry, hibe_cmp)extern const struct rb_type *const hiballoc_addr_RBT_TYPE; __attribute__
((__unused__)) static inline void hiballoc_addr_RBT_INIT(struct
hiballoc_addr *head) { _rb_init(&head->rbh_root); } __attribute__
((__unused__)) static inline struct hiballoc_entry * hiballoc_addr_RBT_INSERT
(struct hiballoc_addr *head, struct hiballoc_entry *elm) { return
_rb_insert(hiballoc_addr_RBT_TYPE, &head->rbh_root, elm
); } __attribute__((__unused__)) static inline struct hiballoc_entry
* hiballoc_addr_RBT_REMOVE(struct hiballoc_addr *head, struct
hiballoc_entry *elm) { return _rb_remove(hiballoc_addr_RBT_TYPE
, &head->rbh_root, elm); } __attribute__((__unused__))
static inline struct hiballoc_entry * hiballoc_addr_RBT_FIND
(struct hiballoc_addr *head, const struct hiballoc_entry *key
) { return _rb_find(hiballoc_addr_RBT_TYPE, &head->rbh_root
, key); } __attribute__((__unused__)) static inline struct hiballoc_entry
* hiballoc_addr_RBT_NFIND(struct hiballoc_addr *head, const struct
hiballoc_entry *key) { return _rb_nfind(hiballoc_addr_RBT_TYPE
, &head->rbh_root, key); } __attribute__((__unused__))
static inline struct hiballoc_entry * hiballoc_addr_RBT_ROOT
(struct hiballoc_addr *head) { return _rb_root(hiballoc_addr_RBT_TYPE
, &head->rbh_root); } __attribute__((__unused__)) static
inline int hiballoc_addr_RBT_EMPTY(struct hiballoc_addr *head
) { return _rb_empty(&head->rbh_root); } __attribute__
((__unused__)) static inline struct hiballoc_entry * hiballoc_addr_RBT_MIN
(struct hiballoc_addr *head) { return _rb_min(hiballoc_addr_RBT_TYPE
, &head->rbh_root); } __attribute__((__unused__)) static
inline struct hiballoc_entry * hiballoc_addr_RBT_MAX(struct hiballoc_addr
*head) { return _rb_max(hiballoc_addr_RBT_TYPE, &head->
rbh_root); } __attribute__((__unused__)) static inline struct
hiballoc_entry * hiballoc_addr_RBT_NEXT(struct hiballoc_entry
*elm) { return _rb_next(hiballoc_addr_RBT_TYPE, elm); } __attribute__
((__unused__)) static inline struct hiballoc_entry * hiballoc_addr_RBT_PREV
(struct hiballoc_entry *elm) { return _rb_prev(hiballoc_addr_RBT_TYPE
, elm); } __attribute__((__unused__)) static inline struct hiballoc_entry
* hiballoc_addr_RBT_LEFT(struct hiballoc_entry *elm) { return
_rb_left(hiballoc_addr_RBT_TYPE, elm); } __attribute__((__unused__
)) static inline struct hiballoc_entry * hiballoc_addr_RBT_RIGHT
(struct hiballoc_entry *elm) { return _rb_right(hiballoc_addr_RBT_TYPE
, elm); } __attribute__((__unused__)) static inline struct hiballoc_entry
* hiballoc_addr_RBT_PARENT(struct hiballoc_entry *elm) { return
_rb_parent(hiballoc_addr_RBT_TYPE, elm); } __attribute__((__unused__
)) static inline void hiballoc_addr_RBT_SET_LEFT(struct hiballoc_entry
*elm, struct hiballoc_entry *left) { _rb_set_left(hiballoc_addr_RBT_TYPE
, elm, left); } __attribute__((__unused__)) static inline void
hiballoc_addr_RBT_SET_RIGHT(struct hiballoc_entry *elm, struct
hiballoc_entry *right) { _rb_set_right(hiballoc_addr_RBT_TYPE
, elm, right); } __attribute__((__unused__)) static inline void
hiballoc_addr_RBT_SET_PARENT(struct hiballoc_entry *elm, struct
hiballoc_entry *parent) { _rb_set_parent(hiballoc_addr_RBT_TYPE
, elm, parent); } __attribute__((__unused__)) static inline void
hiballoc_addr_RBT_POISON(struct hiballoc_entry *elm, unsigned
long poison) { _rb_poison(hiballoc_addr_RBT_TYPE, elm, poison
); } __attribute__((__unused__)) static inline int hiballoc_addr_RBT_CHECK
(struct hiballoc_entry *elm, unsigned long poison) { return _rb_check
(hiballoc_addr_RBT_TYPE, elm, poison); }
170
171/*
172 * Given a hiballoc entry, return the address it manages.
173 */
174static __inline void *
175hib_entry_to_addr(struct hiballoc_entry *entry)
176{
177 caddr_t addr;
178
179 addr = (caddr_t)entry;
180 addr += HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8));
181 return addr;
182}
183
184/*
185 * Given an address, find the hiballoc that corresponds.
186 */
187static __inline struct hiballoc_entry*
188hib_addr_to_entry(void *addr_param)
189{
190 caddr_t addr;
191
192 addr = (caddr_t)addr_param;
193 addr -= HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8));
194 return (struct hiballoc_entry*)addr;
195}
196
197RBT_GENERATE(hiballoc_addr, hiballoc_entry, hibe_entry, hibe_cmp)static int hiballoc_addr_RBT_COMPARE(const void *lptr, const void
*rptr) { const struct hiballoc_entry *l = lptr, *r = rptr; return
hibe_cmp(l, r); } static const struct rb_type hiballoc_addr_RBT_INFO
= { hiballoc_addr_RBT_COMPARE, ((void *)0), __builtin_offsetof
(struct hiballoc_entry, hibe_entry), }; const struct rb_type *
const hiballoc_addr_RBT_TYPE = &hiballoc_addr_RBT_INFO
;
198
199/*
200 * Allocate memory from the arena.
201 *
202 * Returns NULL if no memory is available.
203 */
204void *
205hib_alloc(struct hiballoc_arena *arena, size_t alloc_sz)
206{
207 struct hiballoc_entry *entry, *new_entry;
208 size_t find_sz;
209
210 /*
211 * Enforce alignment of HIB_ALIGN bytes.
212 *
213 * Note that, because the entry is put in front of the allocation,
214 * 0-byte allocations are guaranteed a unique address.
215 */
216 alloc_sz = roundup(alloc_sz, HIB_ALIGN)((((alloc_sz)+((8)-1))/(8))*(8));
217
218 /*
219 * Find an entry with hibe_space >= find_sz.
220 *
221 * If the root node is not large enough, we switch to tree traversal.
222 * Because all entries are made at the bottom of the free space,
223 * traversal from the end has a slightly better chance of yielding
224 * a sufficiently large space.
225 */
226 find_sz = alloc_sz + HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8));
227 entry = RBT_ROOT(hiballoc_addr, &arena->hib_addrs)hiballoc_addr_RBT_ROOT(&arena->hib_addrs);
228 if (entry != NULL((void *)0) && entry->hibe_space < find_sz) {
229 RBT_FOREACH_REVERSE(entry, hiballoc_addr, &arena->hib_addrs)for ((entry) = hiballoc_addr_RBT_MAX((&arena->hib_addrs
)); (entry) != ((void *)0); (entry) = hiballoc_addr_RBT_PREV(
(entry)))
{
230 if (entry->hibe_space >= find_sz)
231 break;
232 }
233 }
234
235 /*
236 * Insufficient or too fragmented memory.
237 */
238 if (entry == NULL((void *)0))
239 return NULL((void *)0);
240
241 /*
242 * Create new entry in allocated space.
243 */
244 new_entry = (struct hiballoc_entry*)(
245 (caddr_t)hib_entry_to_addr(entry) + entry->hibe_use);
246 new_entry->hibe_space = entry->hibe_space - find_sz;
247 new_entry->hibe_use = alloc_sz;
248
249 /*
250 * Insert entry.
251 */
252 if (RBT_INSERT(hiballoc_addr, &arena->hib_addrs, new_entry)hiballoc_addr_RBT_INSERT(&arena->hib_addrs, new_entry) != NULL((void *)0))
253 panic("hib_alloc: insert failure");
254 entry->hibe_space = 0;
255
256 /* Return address managed by entry. */
257 return hib_entry_to_addr(new_entry);
258}
259
260void
261hib_getentropy(char **bufp, size_t *bufplen)
262{
263 if (!bufp || !bufplen)
264 return;
265
266 *bufp = (char *)(global_piglet_va + (29 * PAGE_SIZE(1 << 12)));
267 *bufplen = PAGE_SIZE(1 << 12);
268}
269
270/*
271 * Free a pointer previously allocated from this arena.
272 *
273 * If addr is NULL, this will be silently accepted.
274 */
275void
276hib_free(struct hiballoc_arena *arena, void *addr)
277{
278 struct hiballoc_entry *entry, *prev;
279
280 if (addr == NULL((void *)0))
281 return;
282
283 /*
284 * Derive entry from addr and check it is really in this arena.
285 */
286 entry = hib_addr_to_entry(addr);
287 if (RBT_FIND(hiballoc_addr, &arena->hib_addrs, entry)hiballoc_addr_RBT_FIND(&arena->hib_addrs, entry) != entry)
288 panic("hib_free: freed item %p not in hib arena", addr);
289
290 /*
291 * Give the space in entry to its predecessor.
292 *
293 * If entry has no predecessor, change its used space into free space
294 * instead.
295 */
296 prev = RBT_PREV(hiballoc_addr, entry)hiballoc_addr_RBT_PREV(entry);
297 if (prev != NULL((void *)0) &&
298 (void *)((caddr_t)prev + HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8)) +
299 prev->hibe_use + prev->hibe_space) == entry) {
300 /* Merge entry. */
301 RBT_REMOVE(hiballoc_addr, &arena->hib_addrs, entry)hiballoc_addr_RBT_REMOVE(&arena->hib_addrs, entry);
302 prev->hibe_space += HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8)) +
303 entry->hibe_use + entry->hibe_space;
304 } else {
305 /* Flip used memory to free space. */
306 entry->hibe_space += entry->hibe_use;
307 entry->hibe_use = 0;
308 }
309}
310
311/*
312 * Initialize hiballoc.
313 *
314 * The allocator will manage memory at ptr, which is len bytes.
315 */
316int
317hiballoc_init(struct hiballoc_arena *arena, void *p_ptr, size_t p_len)
318{
319 struct hiballoc_entry *entry;
320 caddr_t ptr;
321 size_t len;
322
323 RBT_INIT(hiballoc_addr, &arena->hib_addrs)hiballoc_addr_RBT_INIT(&arena->hib_addrs);
324
325 /*
326 * Hib allocator enforces HIB_ALIGN alignment.
327 * Fixup ptr and len.
328 */
329 ptr = (caddr_t)roundup((vaddr_t)p_ptr, HIB_ALIGN)(((((vaddr_t)p_ptr)+((8)-1))/(8))*(8));
330 len = p_len - ((size_t)ptr - (size_t)p_ptr);
331 len &= ~((size_t)HIB_ALIGN8 - 1);
332
333 /*
334 * Insufficient memory to be able to allocate and also do bookkeeping.
335 */
336 if (len <= HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8)))
337 return ENOMEM12;
338
339 /*
340 * Create entry describing space.
341 */
342 entry = (struct hiballoc_entry*)ptr;
343 entry->hibe_use = 0;
344 entry->hibe_space = len - HIB_SIZEOF(struct hiballoc_entry)((((sizeof(struct hiballoc_entry))+((8)-1))/(8))*(8));
345 RBT_INSERT(hiballoc_addr, &arena->hib_addrs, entry)hiballoc_addr_RBT_INSERT(&arena->hib_addrs, entry);
346
347 return 0;
348}
349
350/*
351 * Zero all free memory.
352 */
353void
354uvm_pmr_zero_everything(void)
355{
356 struct uvm_pmemrange *pmr;
357 struct vm_page *pg;
358 int i;
359
360 uvm_lock_fpageq()mtx_enter(&uvm.fpageqlock);
361 TAILQ_FOREACH(pmr, &uvm.pmr_control.use, pmr_use)for((pmr) = ((&uvm.pmr_control.use)->tqh_first); (pmr)
!= ((void *)0); (pmr) = ((pmr)->pmr_use.tqe_next))
{
362 /* Zero single pages. */
363 while ((pg = TAILQ_FIRST(&pmr->single[UVM_PMR_MEMTYPE_DIRTY])((&pmr->single[0])->tqh_first))
364 != NULL((void *)0)) {
365 uvm_pmr_remove(pmr, pg);
366 uvm_pagezero(pg);
367 atomic_setbits_intx86_atomic_setbits_u32(&pg->pg_flags, PG_ZERO0x00000100);
368 uvmexp.zeropages++;
369 uvm_pmr_insert(pmr, pg, 0);
370 }
371
372 /* Zero multi page ranges. */
373 while ((pg = RBT_ROOT(uvm_pmr_size,uvm_pmr_size_RBT_ROOT(&pmr->size[0])
374 &pmr->size[UVM_PMR_MEMTYPE_DIRTY])uvm_pmr_size_RBT_ROOT(&pmr->size[0])) != NULL((void *)0)) {
375 pg--; /* Size tree always has second page. */
376 uvm_pmr_remove(pmr, pg);
377 for (i = 0; i < pg->fpgsz; i++) {
378 uvm_pagezero(&pg[i]);
379 atomic_setbits_intx86_atomic_setbits_u32(&pg[i].pg_flags, PG_ZERO0x00000100);
380 uvmexp.zeropages++;
381 }
382 uvm_pmr_insert(pmr, pg, 0);
383 }
384 }
385 uvm_unlock_fpageq()mtx_leave(&uvm.fpageqlock);
386}
387
388/*
389 * Mark all memory as dirty.
390 *
391 * Used to inform the system that the clean memory isn't clean for some
392 * reason, for example because we just came back from hibernate.
393 */
394void
395uvm_pmr_dirty_everything(void)
396{
397 struct uvm_pmemrange *pmr;
398 struct vm_page *pg;
399 int i;
400
401 uvm_lock_fpageq()mtx_enter(&uvm.fpageqlock);
402 TAILQ_FOREACH(pmr, &uvm.pmr_control.use, pmr_use)for((pmr) = ((&uvm.pmr_control.use)->tqh_first); (pmr)
!= ((void *)0); (pmr) = ((pmr)->pmr_use.tqe_next))
{
403 /* Dirty single pages. */
404 while ((pg = TAILQ_FIRST(&pmr->single[UVM_PMR_MEMTYPE_ZERO])((&pmr->single[1])->tqh_first))
405 != NULL((void *)0)) {
406 uvm_pmr_remove(pmr, pg);
407 atomic_clearbits_intx86_atomic_clearbits_u32(&pg->pg_flags, PG_ZERO0x00000100);
408 uvm_pmr_insert(pmr, pg, 0);
409 }
410
411 /* Dirty multi page ranges. */
412 while ((pg = RBT_ROOT(uvm_pmr_size,uvm_pmr_size_RBT_ROOT(&pmr->size[1])
413 &pmr->size[UVM_PMR_MEMTYPE_ZERO])uvm_pmr_size_RBT_ROOT(&pmr->size[1])) != NULL((void *)0)) {
414 pg--; /* Size tree always has second page. */
415 uvm_pmr_remove(pmr, pg);
416 for (i = 0; i < pg->fpgsz; i++)
417 atomic_clearbits_intx86_atomic_clearbits_u32(&pg[i].pg_flags, PG_ZERO0x00000100);
418 uvm_pmr_insert(pmr, pg, 0);
419 }
420 }
421
422 uvmexp.zeropages = 0;
423 uvm_unlock_fpageq()mtx_leave(&uvm.fpageqlock);
424}
425
426/*
427 * Allocate an area that can hold sz bytes and doesn't overlap with
428 * the piglet at piglet_pa.
429 */
430int
431uvm_pmr_alloc_pig(paddr_t *pa, psize_t sz, paddr_t piglet_pa)
432{
433 struct uvm_constraint_range pig_constraint;
434 struct kmem_pa_mode kp_pig = {
435 .kp_constraint = &pig_constraint,
436 .kp_maxseg = 1
437 };
438 vaddr_t va;
439
440 sz = round_page(sz)(((sz) + ((1 << 12) - 1)) & ~((1 << 12) - 1));
441
442 pig_constraint.ucr_low = piglet_pa + 4 * HIBERNATE_CHUNK_SIZE0x400000;
443 pig_constraint.ucr_high = -1;
444
445 va = (vaddr_t)km_alloc(sz, &kv_any, &kp_pig, &kd_nowait);
446 if (va == 0) {
447 pig_constraint.ucr_low = 0;
448 pig_constraint.ucr_high = piglet_pa - 1;
449
450 va = (vaddr_t)km_alloc(sz, &kv_any, &kp_pig, &kd_nowait);
451 if (va == 0)
452 return ENOMEM12;
453 }
454
455 pmap_extract(pmap_kernel()(&kernel_pmap_store), va, pa);
456 return 0;
457}
458
459/*
460 * Allocate a piglet area.
461 *
462 * This needs to be in DMA-safe memory.
463 * Piglets are aligned.
464 *
465 * sz and align in bytes.
466 *
467 * The call will sleep for the pagedaemon to attempt to free memory.
468 * The pagedaemon may decide its not possible to free enough memory, causing
469 * the allocation to fail.
470 */
471int
472uvm_pmr_alloc_piglet(vaddr_t *va, paddr_t *pa, vsize_t sz, paddr_t align)
473{
474 struct kmem_pa_mode kp_piglet = {
475 .kp_constraint = &dma_constraint,
476 .kp_align = align,
477 .kp_maxseg = 1
478 };
479
480 /* Ensure align is a power of 2 */
481 KASSERT((align & (align - 1)) == 0)(((align & (align - 1)) == 0) ? (void)0 : __assert("diagnostic "
, "/usr/src/sys/kern/subr_hibernate.c", 481, "(align & (align - 1)) == 0"
))
;
482
483 /*
484 * Fixup arguments: align must be at least PAGE_SIZE,
485 * sz will be converted to pagecount, since that is what
486 * pmemrange uses internally.
487 */
488 if (align < PAGE_SIZE(1 << 12))
489 kp_piglet.kp_align = PAGE_SIZE(1 << 12);
490
491 sz = round_page(sz)(((sz) + ((1 << 12) - 1)) & ~((1 << 12) - 1));
492
493 *va = (vaddr_t)km_alloc(sz, &kv_any, &kp_piglet, &kd_nowait);
494 if (*va == 0)
495 return ENOMEM12;
496
497 pmap_extract(pmap_kernel()(&kernel_pmap_store), *va, pa);
498 return 0;
499}
500
501/*
502 * Free a piglet area.
503 */
504void
505uvm_pmr_free_piglet(vaddr_t va, vsize_t sz)
506{
507 /*
508 * Fix parameters.
509 */
510 sz = round_page(sz)(((sz) + ((1 << 12) - 1)) & ~((1 << 12) - 1));
511
512 /*
513 * Free the physical and virtual memory.
514 */
515 km_free((void *)va, sz, &kv_any, &kp_dma_contig);
516}
517
518/*
519 * Physmem RLE compression support.
520 *
521 * Given a physical page address, return the number of pages starting at the
522 * address that are free. Clamps to the number of pages in
523 * HIBERNATE_CHUNK_SIZE. Returns 0 if the page at addr is not free.
524 */
525int
526uvm_page_rle(paddr_t addr)
527{
528 struct vm_page *pg, *pg_end;
529 struct vm_physseg *vmp;
530 int pseg_idx, off_idx;
531
532 pseg_idx = vm_physseg_find(atop(addr)((addr) >> 12), &off_idx);
533 if (pseg_idx == -1)
534 return 0;
535
536 vmp = &vm_physmem[pseg_idx];
537 pg = &vmp->pgs[off_idx];
538 if (!(pg->pg_flags & PQ_FREE0x00010000))
539 return 0;
540
541 /*
542 * Search for the first non-free page after pg.
543 * Note that the page may not be the first page in a free pmemrange,
544 * therefore pg->fpgsz cannot be used.
545 */
546 for (pg_end = pg; pg_end <= vmp->lastpg &&
547 (pg_end->pg_flags & PQ_FREE0x00010000) == PQ_FREE0x00010000 &&
548 (pg_end - pg) < HIBERNATE_CHUNK_SIZE0x400000/PAGE_SIZE(1 << 12); pg_end++)
549 ;
550 return pg_end - pg;
551}
552
553/*
554 * Calculate a hopefully unique version # for this kernel, based upon
555 * how it was linked.
556 */
557u_int32_t
558hibsum(void)
559{
560 return ((long)malloc ^ (long)km_alloc ^ (long)printf ^ (long)strlen);
561}
562
563
564/*
565 * Fills out the hibernate_info union pointed to by hib
566 * with information about this machine (swap signature block
567 * offsets, number of memory ranges, kernel in use, etc)
568 */
569int
570get_hibernate_info(union hibernate_info *hib, int suspend)
571{
572 struct disklabel dl;
573 char err_string[128], *dl_ret;
574 int part;
575
576#ifndef NO_PROPOLICE
577 /* Save propolice guard */
578 hib->guard = __guard_local;
579#endif /* ! NO_PROPOLICE */
580
581 /* Determine I/O function to use */
582 hib->io_func = get_hibernate_io_function(swdevt[0].sw_dev);
583 if (hib->io_func == NULL((void *)0))
584 return (1);
585
586 /* Calculate hibernate device */
587 hib->dev = swdevt[0].sw_dev;
588
589 /* Read disklabel (used to calculate signature and image offsets) */
590 dl_ret = disk_readlabel(&dl, hib->dev, err_string, sizeof(err_string));
591
592 if (dl_ret) {
593 printf("Hibernate error reading disklabel: %s\n", dl_ret);
594 return (1);
595 }
596
597 /* Make sure we have a swap partition. */
598 part = DISKPART(hib->dev)(((unsigned)((hib->dev) & 0xff) | (((hib->dev) &
0xffff0000) >> 8)) % 16)
;
599 if (dl.d_npartitions <= part ||
600 dl.d_partitions[part].p_fstype != FS_SWAP1 ||
601 DL_GETPSIZE(&dl.d_partitions[part])(((u_int64_t)(&dl.d_partitions[part])->p_sizeh <<
32) + (&dl.d_partitions[part])->p_size)
== 0)
602 return (1);
603
604 /* Magic number */
605 hib->magic = HIBERNATE_MAGIC0x0B5D0B5D;
606
607 /* Calculate signature block location */
608 hib->sig_offset = DL_GETPSIZE(&dl.d_partitions[part])(((u_int64_t)(&dl.d_partitions[part])->p_sizeh <<
32) + (&dl.d_partitions[part])->p_size)
-
609 sizeof(union hibernate_info)/DEV_BSIZE(1 << 9);
610
611 /* Stash kernel version information */
612 memset(&hib->kernel_version, 0, 128)__builtin_memset((&hib->kernel_version), (0), (128));
613 bcopy(version, &hib->kernel_version,
614 min(strlen(version), sizeof(hib->kernel_version)-1));
615 hib->kernel_sum = hibsum();
616
617 if (suspend) {
618 /* Grab the previously-allocated piglet addresses */
619 hib->piglet_va = global_piglet_va;
620 hib->piglet_pa = global_piglet_pa;
621 hib->io_page = (void *)hib->piglet_va;
622
623 /*
624 * Initialization of the hibernate IO function for drivers
625 * that need to do prep work (such as allocating memory or
626 * setting up data structures that cannot safely be done
627 * during suspend without causing side effects). There is
628 * a matching HIB_DONE call performed after the write is
629 * completed.
630 */
631 if (hib->io_func(hib->dev, DL_GETPOFFSET(&dl.d_partitions[part])(((u_int64_t)(&dl.d_partitions[part])->p_offseth <<
32) + (&dl.d_partitions[part])->p_offset)
,
632 (vaddr_t)NULL((void *)0), DL_GETPSIZE(&dl.d_partitions[part])(((u_int64_t)(&dl.d_partitions[part])->p_sizeh <<
32) + (&dl.d_partitions[part])->p_size)
,
633 HIB_INIT-1, hib->io_page))
634 goto fail;
635
636 } else {
637 /*
638 * Resuming kernels use a regular private page for the driver
639 * No need to free this I/O page as it will vanish as part of
640 * the resume.
641 */
642 hib->io_page = malloc(PAGE_SIZE(1 << 12), M_DEVBUF2, M_NOWAIT0x0002);
643 if (!hib->io_page)
644 goto fail;
645 }
646
647 if (get_hibernate_info_md(hib))
648 goto fail;
649
650 return (0);
651
652fail:
653 return (1);
654}
655
656/*
657 * Allocate nitems*size bytes from the hiballoc area presently in use
658 */
659void *
660hibernate_zlib_alloc(void *unused, int nitems, int size)
661{
662 struct hibernate_zlib_state *hibernate_state;
663
664 hibernate_state =
665 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
666
667 return hib_alloc(&hibernate_state->hiballoc_arena, nitems*size);
668}
669
670/*
671 * Free the memory pointed to by addr in the hiballoc area presently in
672 * use
673 */
674void
675hibernate_zlib_free(void *unused, void *addr)
676{
677 struct hibernate_zlib_state *hibernate_state;
678
679 hibernate_state =
680 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
681
682 hib_free(&hibernate_state->hiballoc_arena, addr);
683}
684
685/*
686 * Inflate next page of data from the image stream.
687 * The rle parameter is modified on exit to contain the number of pages to
688 * skip in the output stream (or 0 if this page was inflated into).
689 *
690 * Returns 0 if the stream contains additional data, or 1 if the stream is
691 * finished.
692 */
693int
694hibernate_inflate_page(int *rle)
695{
696 struct hibernate_zlib_state *hibernate_state;
697 int i;
698
699 hibernate_state =
700 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
701
702 /* Set up the stream for RLE code inflate */
703 hibernate_state->hib_stream.next_out = (unsigned char *)rle;
704 hibernate_state->hib_stream.avail_out = sizeof(*rle);
705
706 /* Inflate RLE code */
707 i = inflate(&hibernate_state->hib_stream, Z_SYNC_FLUSH2);
708 if (i != Z_OK0 && i != Z_STREAM_END1) {
709 /*
710 * XXX - this will likely reboot/hang most machines
711 * since the console output buffer will be unmapped,
712 * but there's not much else we can do here.
713 */
714 panic("rle inflate stream error");
715 }
716
717 if (hibernate_state->hib_stream.avail_out != 0) {
718 /*
719 * XXX - this will likely reboot/hang most machines
720 * since the console output buffer will be unmapped,
721 * but there's not much else we can do here.
722 */
723 panic("rle short inflate error");
724 }
725
726 if (*rle < 0 || *rle > 1024) {
727 /*
728 * XXX - this will likely reboot/hang most machines
729 * since the console output buffer will be unmapped,
730 * but there's not much else we can do here.
731 */
732 panic("invalid rle count");
733 }
734
735 if (i == Z_STREAM_END1)
736 return (1);
737
738 if (*rle != 0)
739 return (0);
740
741 /* Set up the stream for page inflate */
742 hibernate_state->hib_stream.next_out =
743 (unsigned char *)HIBERNATE_INFLATE_PAGE((1 << 12) * 33);
744 hibernate_state->hib_stream.avail_out = PAGE_SIZE(1 << 12);
745
746 /* Process next block of data */
747 i = inflate(&hibernate_state->hib_stream, Z_SYNC_FLUSH2);
748 if (i != Z_OK0 && i != Z_STREAM_END1) {
749 /*
750 * XXX - this will likely reboot/hang most machines
751 * since the console output buffer will be unmapped,
752 * but there's not much else we can do here.
753 */
754 panic("inflate error");
755 }
756
757 /* We should always have extracted a full page ... */
758 if (hibernate_state->hib_stream.avail_out != 0) {
759 /*
760 * XXX - this will likely reboot/hang most machines
761 * since the console output buffer will be unmapped,
762 * but there's not much else we can do here.
763 */
764 panic("incomplete page");
765 }
766
767 return (i == Z_STREAM_END1);
768}
769
770/*
771 * Inflate size bytes from src into dest, skipping any pages in
772 * [src..dest] that are special (see hibernate_inflate_skip)
773 *
774 * This function executes while using the resume-time stack
775 * and pmap, and therefore cannot use ddb/printf/etc. Doing so
776 * will likely hang or reset the machine since the console output buffer
777 * will be unmapped.
778 */
779void
780hibernate_inflate_region(union hibernate_info *hib, paddr_t dest,
781 paddr_t src, size_t size)
782{
783 int end_stream = 0, rle, skip;
784 struct hibernate_zlib_state *hibernate_state;
785
786 hibernate_state =
787 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
788
789 hibernate_state->hib_stream.next_in = (unsigned char *)src;
790 hibernate_state->hib_stream.avail_in = size;
791
792 do {
793 /*
794 * Is this a special page? If yes, redirect the
795 * inflate output to a scratch page (eg, discard it)
796 */
797 skip = hibernate_inflate_skip(hib, dest);
798 if (skip == HIB_SKIP1) {
799 hibernate_enter_resume_mapping(
800 HIBERNATE_INFLATE_PAGE((1 << 12) * 33),
801 HIBERNATE_INFLATE_PAGE((1 << 12) * 33), 0);
802 } else if (skip == HIB_MOVE2) {
803 /*
804 * Special case : retguard region. This gets moved
805 * temporarily into the piglet region and copied into
806 * place immediately before resume
807 */
808 hibernate_enter_resume_mapping(
809 HIBERNATE_INFLATE_PAGE((1 << 12) * 33),
810 hib->piglet_pa + (110 * PAGE_SIZE(1 << 12)) +
811 hib->retguard_ofs, 0);
812 hib->retguard_ofs += PAGE_SIZE(1 << 12);
813 if (hib->retguard_ofs > 255 * PAGE_SIZE(1 << 12)) {
814 /*
815 * XXX - this will likely reboot/hang most
816 * machines since the console output
817 * buffer will be unmapped, but there's
818 * not much else we can do here.
819 */
820 panic("retguard move error, out of space");
821 }
822 } else {
823 hibernate_enter_resume_mapping(
824 HIBERNATE_INFLATE_PAGE((1 << 12) * 33), dest, 0);
825 }
826
827 hibernate_flush();
828 end_stream = hibernate_inflate_page(&rle);
829
830 if (rle == 0)
831 dest += PAGE_SIZE(1 << 12);
832 else
833 dest += (rle * PAGE_SIZE(1 << 12));
834 } while (!end_stream);
835}
836
837/*
838 * deflate from src into the I/O page, up to 'remaining' bytes
839 *
840 * Returns number of input bytes consumed, and may reset
841 * the 'remaining' parameter if not all the output space was consumed
842 * (this information is needed to know how much to write to disk)
843 */
844size_t
845hibernate_deflate(union hibernate_info *hib, paddr_t src,
846 size_t *remaining)
847{
848 vaddr_t hibernate_io_page = hib->piglet_va + PAGE_SIZE(1 << 12);
849 struct hibernate_zlib_state *hibernate_state;
850
851 hibernate_state =
852 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
853
854 /* Set up the stream for deflate */
855 hibernate_state->hib_stream.next_in = (unsigned char *)src;
856 hibernate_state->hib_stream.avail_in = PAGE_SIZE(1 << 12) - (src & PAGE_MASK((1 << 12) - 1));
857 hibernate_state->hib_stream.next_out =
858 (unsigned char *)hibernate_io_page + (PAGE_SIZE(1 << 12) - *remaining);
859 hibernate_state->hib_stream.avail_out = *remaining;
860
861 /* Process next block of data */
862 if (deflate(&hibernate_state->hib_stream, Z_SYNC_FLUSH2) != Z_OK0)
863 panic("hibernate zlib deflate error");
864
865 /* Update pointers and return number of bytes consumed */
866 *remaining = hibernate_state->hib_stream.avail_out;
867 return (PAGE_SIZE(1 << 12) - (src & PAGE_MASK((1 << 12) - 1))) -
868 hibernate_state->hib_stream.avail_in;
869}
870
871/*
872 * Write the hibernation information specified in hiber_info
873 * to the location in swap previously calculated (last block of
874 * swap), called the "signature block".
875 */
876int
877hibernate_write_signature(union hibernate_info *hib)
878{
879 /* Write hibernate info to disk */
880 return (hib->io_func(hib->dev, hib->sig_offset,
881 (vaddr_t)hib, DEV_BSIZE(1 << 9), HIB_W1,
882 hib->io_page));
883}
884
885/*
886 * Write the memory chunk table to the area in swap immediately
887 * preceding the signature block. The chunk table is stored
888 * in the piglet when this function is called. Returns errno.
889 */
890int
891hibernate_write_chunktable(union hibernate_info *hib)
892{
893 vaddr_t hibernate_chunk_table_start;
894 size_t hibernate_chunk_table_size;
895 int i, err;
896
897 hibernate_chunk_table_size = HIBERNATE_CHUNK_TABLE_SIZE0x100000;
898
899 hibernate_chunk_table_start = hib->piglet_va +
900 HIBERNATE_CHUNK_SIZE0x400000;
901
902 /* Write chunk table */
903 for (i = 0; i < hibernate_chunk_table_size; i += MAXPHYS(64 * 1024)) {
904 if ((err = hib->io_func(hib->dev,
905 hib->chunktable_offset + (i/DEV_BSIZE(1 << 9)),
906 (vaddr_t)(hibernate_chunk_table_start + i),
907 MAXPHYS(64 * 1024), HIB_W1, hib->io_page))) {
908 DPRINTF("chunktable write error: %d\n", err);
909 return (err);
910 }
911 }
912
913 return (0);
914}
915
916/*
917 * Write an empty hiber_info to the swap signature block, which is
918 * guaranteed to not match any valid hib.
919 */
920int
921hibernate_clear_signature(union hibernate_info *hib)
922{
923 union hibernate_info blank_hiber_info;
924
925 /* Zero out a blank hiber_info */
926 memset(&blank_hiber_info, 0, sizeof(union hibernate_info))__builtin_memset((&blank_hiber_info), (0), (sizeof(union hibernate_info
)))
;
927
928 /* Write (zeroed) hibernate info to disk */
929 DPRINTF("clearing hibernate signature block location: %lld\n",
930 hib->sig_offset);
931 if (hibernate_block_io(hib,
932 hib->sig_offset,
933 DEV_BSIZE(1 << 9), (vaddr_t)&blank_hiber_info, 1))
934 printf("Warning: could not clear hibernate signature\n");
935
936 return (0);
937}
938
939/*
940 * Compare two hibernate_infos to determine if they are the same (eg,
941 * we should be performing a hibernate resume on this machine.
942 * Not all fields are checked - just enough to verify that the machine
943 * has the same memory configuration and kernel as the one that
944 * wrote the signature previously.
945 */
946int
947hibernate_compare_signature(union hibernate_info *mine,
948 union hibernate_info *disk)
949{
950 u_int i;
951
952 if (mine->nranges != disk->nranges) {
953 printf("unhibernate failed: memory layout changed\n");
954 return (1);
955 }
956
957 if (strcmp(mine->kernel_version, disk->kernel_version) != 0) {
958 printf("unhibernate failed: original kernel changed\n");
959 return (1);
960 }
961
962 if (hibsum() != disk->kernel_sum) {
963 printf("unhibernate failed: original kernel changed\n");
964 return (1);
965 }
966
967 for (i = 0; i < mine->nranges; i++) {
968 if ((mine->ranges[i].base != disk->ranges[i].base) ||
969 (mine->ranges[i].end != disk->ranges[i].end) ) {
970 DPRINTF("hib range %d mismatch [%p-%p != %p-%p]\n",
971 i,
972 (void *)mine->ranges[i].base,
973 (void *)mine->ranges[i].end,
974 (void *)disk->ranges[i].base,
975 (void *)disk->ranges[i].end);
976 printf("unhibernate failed: memory size changed\n");
977 return (1);
978 }
979 }
980
981 return (0);
982}
983
984/*
985 * Transfers xfer_size bytes between the hibernate device specified in
986 * hib_info at offset blkctr and the vaddr specified at dest.
987 *
988 * Separate offsets and pages are used to handle misaligned reads (reads
989 * that span a page boundary).
990 *
991 * blkctr specifies a relative offset (relative to the start of swap),
992 * not an absolute disk offset
993 *
994 */
995int
996hibernate_block_io(union hibernate_info *hib, daddr_t blkctr,
997 size_t xfer_size, vaddr_t dest, int iswrite)
998{
999 struct buf *bp;
1000 struct bdevsw *bdsw;
1001 int error;
1002
1003 bp = geteblk(xfer_size);
1004 bdsw = &bdevsw[major(hib->dev)(((unsigned)(hib->dev) >> 8) & 0xff)];
1005
1006 error = (*bdsw->d_open)(hib->dev, FREAD0x0001, S_IFCHR0020000, curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1007 if (error) {
1008 printf("hibernate_block_io open failed\n");
1009 return (1);
1010 }
1011
1012 if (iswrite)
1013 bcopy((caddr_t)dest, bp->b_data, xfer_size);
1014
1015 bp->b_bcount = xfer_size;
1016 bp->b_blkno = blkctr;
1017 CLR(bp->b_flags, B_READ | B_WRITE | B_DONE)((bp->b_flags) &= ~(0x00008000 | 0x00000000 | 0x00000100
))
;
1018 SET(bp->b_flags, B_BUSY | (iswrite ? B_WRITE : B_READ) | B_RAW)((bp->b_flags) |= (0x00000010 | (iswrite ? 0x00000000 : 0x00008000
) | 0x00004000))
;
1019 bp->b_dev = hib->dev;
1020 (*bdsw->d_strategy)(bp);
1021
1022 error = biowait(bp);
1023 if (error) {
1024 printf("hib block_io biowait error %d blk %lld size %zu\n",
1025 error, (long long)blkctr, xfer_size);
1026 error = (*bdsw->d_close)(hib->dev, 0, S_IFCHR0020000,
1027 curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1028 if (error)
1029 printf("hibernate_block_io error close failed\n");
1030 return (1);
1031 }
1032
1033 error = (*bdsw->d_close)(hib->dev, FREAD0x0001, S_IFCHR0020000, curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1034 if (error) {
1035 printf("hibernate_block_io close failed\n");
1036 return (1);
1037 }
1038
1039 if (!iswrite)
1040 bcopy(bp->b_data, (caddr_t)dest, xfer_size);
1041
1042 bp->b_flags |= B_INVAL0x00000800;
1043 brelse(bp);
1044
1045 return (0);
1046}
1047
1048/*
1049 * Preserve one page worth of random data, generated from the resuming
1050 * kernel's arc4random. After resume, this preserved entropy can be used
1051 * to further improve the un-hibernated machine's entropy pool. This
1052 * random data is stored in the piglet, which is preserved across the
1053 * unpack operation, and is restored later in the resume process (see
1054 * hib_getentropy)
1055 */
1056void
1057hibernate_preserve_entropy(union hibernate_info *hib)
1058{
1059 void *entropy;
1060
1061 entropy = km_alloc(PAGE_SIZE(1 << 12), &kv_any, &kp_none, &kd_nowait);
1062
1063 if (!entropy)
1064 return;
1065
1066 pmap_activate(curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1067 pmap_kenter_pa((vaddr_t)entropy,
1068 (paddr_t)(hib->piglet_pa + (29 * PAGE_SIZE(1 << 12))),
1069 PROT_READ0x01 | PROT_WRITE0x02);
1070
1071 arc4random_buf((void *)entropy, PAGE_SIZE(1 << 12));
1072 pmap_kremove((vaddr_t)entropy, PAGE_SIZE(1 << 12));
1073 km_free(entropy, PAGE_SIZE(1 << 12), &kv_any, &kp_none);
1074}
1075
1076#ifndef NO_PROPOLICE
1077vaddr_t
1078hibernate_unprotect_ssp(void)
1079{
1080 struct kmem_dyn_mode kd_avoidalias;
1081 vaddr_t va = trunc_page((vaddr_t)&__guard_local)(((vaddr_t)&__guard_local) & ~((1 << 12) - 1));
1082 paddr_t pa;
1083
1084 pmap_extract(pmap_kernel()(&kernel_pmap_store), va, &pa);
1085
1086 memset(&kd_avoidalias, 0, sizeof kd_avoidalias)__builtin_memset((&kd_avoidalias), (0), (sizeof kd_avoidalias
))
;
1087 kd_avoidalias.kd_prefer = pa;
1088 kd_avoidalias.kd_waitok = 1;
1089 va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_any, &kp_none, &kd_avoidalias);
1090 if (!va)
1091 panic("hibernate_unprotect_ssp");
1092
1093 pmap_kenter_pa(va, pa, PROT_READ0x01 | PROT_WRITE0x02);
1094 pmap_update(pmap_kernel());
1095
1096 return va;
1097}
1098
1099void
1100hibernate_reprotect_ssp(vaddr_t va)
1101{
1102 pmap_kremove(va, PAGE_SIZE(1 << 12));
1103 km_free((void *)va, PAGE_SIZE(1 << 12), &kv_any, &kp_none);
1104}
1105#endif /* NO_PROPOLICE */
1106
1107/*
1108 * Reads the signature block from swap, checks against the current machine's
1109 * information. If the information matches, perform a resume by reading the
1110 * saved image into the pig area, and unpacking.
1111 *
1112 * Must be called with interrupts enabled.
1113 */
1114void
1115hibernate_resume(void)
1116{
1117 union hibernate_info hib;
1118 int s;
1119#ifndef NO_PROPOLICE
1120 vsize_t off = (vaddr_t)&__guard_local -
1121 trunc_page((vaddr_t)&__guard_local)(((vaddr_t)&__guard_local) & ~((1 << 12) - 1));
1122 vaddr_t guard_va;
1123#endif
1124
1125 /* Get current running machine's hibernate info */
1126 memset(&hib, 0, sizeof(hib))__builtin_memset((&hib), (0), (sizeof(hib)));
1127 if (get_hibernate_info(&hib, 0)) {
1128 DPRINTF("couldn't retrieve machine's hibernate info\n");
1129 return;
1130 }
1131
1132 /* Read hibernate info from disk */
1133 s = splbio()splraise(0x6);
1134
1135 DPRINTF("reading hibernate signature block location: %lld\n",
1136 hib.sig_offset);
1137
1138 if (hibernate_block_io(&hib,
1139 hib.sig_offset,
1140 DEV_BSIZE(1 << 9), (vaddr_t)&disk_hib, 0)) {
1141 DPRINTF("error in hibernate read");
1142 splx(s)spllower(s);
1143 return;
1144 }
1145
1146 /* Check magic number */
1147 if (disk_hib.magic != HIBERNATE_MAGIC0x0B5D0B5D) {
1148 DPRINTF("wrong magic number in hibernate signature: %x\n",
1149 disk_hib.magic);
1150 splx(s)spllower(s);
1151 return;
1152 }
1153
1154 /*
1155 * We (possibly) found a hibernate signature. Clear signature first,
1156 * to prevent accidental resume or endless resume cycles later.
1157 */
1158 if (hibernate_clear_signature(&hib)) {
1159 DPRINTF("error clearing hibernate signature block\n");
1160 splx(s)spllower(s);
1161 return;
1162 }
1163
1164 /*
1165 * If on-disk and in-memory hibernate signatures match,
1166 * this means we should do a resume from hibernate.
1167 */
1168 if (hibernate_compare_signature(&hib, &disk_hib)) {
1169 DPRINTF("mismatched hibernate signature block\n");
1170 splx(s)spllower(s);
1171 return;
1172 }
1173 disk_hib.dev = hib.dev;
1174
1175#ifdef MULTIPROCESSOR1
1176 /* XXX - if we fail later, we may need to rehatch APs on some archs */
1177 DPRINTF("hibernate: quiescing APs\n");
1178 hibernate_quiesce_cpus();
1179#endif /* MULTIPROCESSOR */
1180
1181 /* Read the image from disk into the image (pig) area */
1182 if (hibernate_read_image(&disk_hib))
1183 goto fail;
1184
1185 DPRINTF("hibernate: quiescing devices\n");
1186 if (config_suspend_all(DVACT_QUIESCE2) != 0)
1187 goto fail;
1188
1189#ifndef NO_PROPOLICE
1190 guard_va = hibernate_unprotect_ssp();
1191#endif /* NO_PROPOLICE */
1192
1193 (void) splhigh()splraise(0xd);
1194 hibernate_disable_intr_machdep();
1195 cold = 1;
1196
1197 DPRINTF("hibernate: suspending devices\n");
1198 if (config_suspend_all(DVACT_SUSPEND3) != 0) {
1199 cold = 0;
1200 hibernate_enable_intr_machdep();
1201#ifndef NO_PROPOLICE
1202 hibernate_reprotect_ssp(guard_va);
1203#endif /* ! NO_PROPOLICE */
1204 goto fail;
1205 }
1206
1207 hibernate_preserve_entropy(&disk_hib);
1208
1209 printf("Unpacking image...\n");
1210
1211 /* Switch stacks */
1212 DPRINTF("hibernate: switching stacks\n");
1213 hibernate_switch_stack_machdep();
1214
1215#ifndef NO_PROPOLICE
1216 /* Start using suspended kernel's propolice guard */
1217 *(long *)(guard_va + off) = disk_hib.guard;
1218 hibernate_reprotect_ssp(guard_va);
1219#endif /* ! NO_PROPOLICE */
1220
1221 /* Unpack and resume */
1222 hibernate_unpack_image(&disk_hib);
1223
1224fail:
1225 splx(s)spllower(s);
1226 printf("\nUnable to resume hibernated image\n");
1227}
1228
1229/*
1230 * Unpack image from pig area to original location by looping through the
1231 * list of output chunks in the order they should be restored (fchunks).
1232 *
1233 * Note that due to the stack smash protector and the fact that we have
1234 * switched stacks, it is not permitted to return from this function.
1235 */
1236void
1237hibernate_unpack_image(union hibernate_info *hib)
1238{
1239 struct hibernate_disk_chunk *chunks;
1240 union hibernate_info local_hib;
1241 paddr_t image_cur = global_pig_start;
1242 short i, *fchunks;
1243 char *pva;
1244
1245 /* Piglet will be identity mapped (VA == PA) */
1246 pva = (char *)hib->piglet_pa;
1247
1248 fchunks = (short *)(pva + (4 * PAGE_SIZE(1 << 12)));
1249
1250 chunks = (struct hibernate_disk_chunk *)(pva + HIBERNATE_CHUNK_SIZE0x400000);
1251
1252 /* Can't use hiber_info that's passed in after this point */
1253 bcopy(hib, &local_hib, sizeof(union hibernate_info));
1254 local_hib.retguard_ofs = 0;
1255
1256 /* VA == PA */
1257 local_hib.piglet_va = local_hib.piglet_pa;
1258
1259 /*
1260 * Point of no return. Once we pass this point, only kernel code can
1261 * be accessed. No global variables or other kernel data structures
1262 * are guaranteed to be coherent after unpack starts.
1263 *
1264 * The image is now in high memory (pig area), we unpack from the pig
1265 * to the correct location in memory. We'll eventually end up copying
1266 * on top of ourself, but we are assured the kernel code here is the
1267 * same between the hibernated and resuming kernel, and we are running
1268 * on our own stack, so the overwrite is ok.
1269 */
1270 DPRINTF("hibernate: activating alt. pagetable and starting unpack\n");
1271 hibernate_activate_resume_pt_machdep();
1272
1273 for (i = 0; i < local_hib.chunk_ctr; i++) {
1274 /* Reset zlib for inflate */
1275 if (hibernate_zlib_reset(&local_hib, 0) != Z_OK0)
1276 panic("hibernate failed to reset zlib for inflate");
1277
1278 hibernate_process_chunk(&local_hib, &chunks[fchunks[i]],
1279 image_cur);
1280
1281 image_cur += chunks[fchunks[i]].compressed_size;
1282
1283 }
1284
1285 /*
1286 * Resume the loaded kernel by jumping to the MD resume vector.
1287 * We won't be returning from this call. We pass the location of
1288 * the retguard save area so the MD code can replace it before
1289 * resuming. See the piglet layout at the top of this file for
1290 * more information on the layout of the piglet area.
1291 *
1292 * We use 'global_piglet_va' here since by the time we are at
1293 * this point, we have already unpacked the image, and we want
1294 * the suspended kernel's view of what the piglet was, before
1295 * suspend occurred (since we will need to use that in the retguard
1296 * copy code in hibernate_resume_machdep.)
1297 */
1298 hibernate_resume_machdep(global_piglet_va + (110 * PAGE_SIZE(1 << 12)));
1299}
1300
1301/*
1302 * Bounce a compressed image chunk to the piglet, entering mappings for the
1303 * copied pages as needed
1304 */
1305void
1306hibernate_copy_chunk_to_piglet(paddr_t img_cur, vaddr_t piglet, size_t size)
1307{
1308 size_t ct, ofs;
1309 paddr_t src = img_cur;
1310 vaddr_t dest = piglet;
1311
1312 /* Copy first partial page */
1313 ct = (PAGE_SIZE(1 << 12)) - (src & PAGE_MASK((1 << 12) - 1));
1314 ofs = (src & PAGE_MASK((1 << 12) - 1));
1315
1316 if (ct < PAGE_SIZE(1 << 12)) {
1317 hibernate_enter_resume_mapping(HIBERNATE_INFLATE_PAGE((1 << 12) * 33),
1318 (src - ofs), 0);
1319 hibernate_flush();
1320 bcopy((caddr_t)(HIBERNATE_INFLATE_PAGE((1 << 12) * 33) + ofs), (caddr_t)dest, ct);
1321 src += ct;
1322 dest += ct;
1323 }
1324
1325 /* Copy remaining pages */
1326 while (src < size + img_cur) {
1327 hibernate_enter_resume_mapping(HIBERNATE_INFLATE_PAGE((1 << 12) * 33), src, 0);
1328 hibernate_flush();
1329 ct = PAGE_SIZE(1 << 12);
1330 bcopy((caddr_t)(HIBERNATE_INFLATE_PAGE((1 << 12) * 33)), (caddr_t)dest, ct);
1331 hibernate_flush();
1332 src += ct;
1333 dest += ct;
1334 }
1335}
1336
1337/*
1338 * Process a chunk by bouncing it to the piglet, followed by unpacking
1339 */
1340void
1341hibernate_process_chunk(union hibernate_info *hib,
1342 struct hibernate_disk_chunk *chunk, paddr_t img_cur)
1343{
1344 char *pva = (char *)hib->piglet_va;
1345
1346 hibernate_copy_chunk_to_piglet(img_cur,
1347 (vaddr_t)(pva + (HIBERNATE_CHUNK_SIZE0x400000 * 2)), chunk->compressed_size);
1348 hibernate_inflate_region(hib, chunk->base,
1349 (vaddr_t)(pva + (HIBERNATE_CHUNK_SIZE0x400000 * 2)),
1350 chunk->compressed_size);
1351}
1352
1353/*
1354 * Calculate RLE component for 'inaddr'. Clamps to max RLE pages between
1355 * inaddr and range_end.
1356 */
1357int
1358hibernate_calc_rle(paddr_t inaddr, paddr_t range_end)
1359{
1360 int rle;
1361
1362 rle = uvm_page_rle(inaddr);
1363 KASSERT(rle >= 0 && rle <= MAX_RLE)((rle >= 0 && rle <= (0x400000 / (1 << 12
))) ? (void)0 : __assert("diagnostic ", "/usr/src/sys/kern/subr_hibernate.c"
, 1363, "rle >= 0 && rle <= MAX_RLE"))
;
1364
1365 /* Clamp RLE to range end */
1366 if (rle > 0 && inaddr + (rle * PAGE_SIZE(1 << 12)) > range_end)
1367 rle = (range_end - inaddr) / PAGE_SIZE(1 << 12);
1368
1369 return (rle);
1370}
1371
1372/*
1373 * Write the RLE byte for page at 'inaddr' to the output stream.
1374 * Returns the number of pages to be skipped at 'inaddr'.
1375 */
1376int
1377hibernate_write_rle(union hibernate_info *hib, paddr_t inaddr,
1378 paddr_t range_end, daddr_t *blkctr,
1379 size_t *out_remaining)
1380{
1381 int rle, err, *rleloc;
1382 struct hibernate_zlib_state *hibernate_state;
1383 vaddr_t hibernate_io_page = hib->piglet_va + PAGE_SIZE(1 << 12);
1384
1385 hibernate_state =
1386 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
1387
1388 rle = hibernate_calc_rle(inaddr, range_end);
1389
1390 rleloc = (int *)hibernate_rle_page + MAX_RLE(0x400000 / (1 << 12)) - 1;
1391 *rleloc = rle;
1392
1393 /* Deflate the RLE byte into the stream */
1394 hibernate_deflate(hib, (paddr_t)rleloc, out_remaining);
1395
1396 /* Did we fill the output page? If so, flush to disk */
1397 if (*out_remaining == 0) {
1398 if ((err = hib->io_func(hib->dev, *blkctr + hib->image_offset,
1399 (vaddr_t)hibernate_io_page, PAGE_SIZE(1 << 12), HIB_W1,
1400 hib->io_page))) {
1401 DPRINTF("hib write error %d\n", err);
1402 return (err);
1403 }
1404
1405 *blkctr += PAGE_SIZE(1 << 12) / DEV_BSIZE(1 << 9);
1406 *out_remaining = PAGE_SIZE(1 << 12);
1407
1408 /* If we didn't deflate the entire RLE byte, finish it now */
1409 if (hibernate_state->hib_stream.avail_in != 0)
1410 hibernate_deflate(hib,
1411 (vaddr_t)hibernate_state->hib_stream.next_in,
1412 out_remaining);
1413 }
1414
1415 return (rle);
1416}
1417
1418/*
1419 * Write a compressed version of this machine's memory to disk, at the
1420 * precalculated swap offset:
1421 *
1422 * end of swap - signature block size - chunk table size - memory size
1423 *
1424 * The function begins by looping through each phys mem range, cutting each
1425 * one into MD sized chunks. These chunks are then compressed individually
1426 * and written out to disk, in phys mem order. Some chunks might compress
1427 * more than others, and for this reason, each chunk's size is recorded
1428 * in the chunk table, which is written to disk after the image has
1429 * properly been compressed and written (in hibernate_write_chunktable).
1430 *
1431 * When this function is called, the machine is nearly suspended - most
1432 * devices are quiesced/suspended, interrupts are off, and cold has
1433 * been set. This means that there can be no side effects once the
1434 * write has started, and the write function itself can also have no
1435 * side effects. This also means no printfs are permitted (since printf
1436 * has side effects.)
1437 *
1438 * Return values :
1439 *
1440 * 0 - success
1441 * EIO - I/O error occurred writing the chunks
1442 * EINVAL - Failed to write a complete range
1443 * ENOMEM - Memory allocation failure during preparation of the zlib arena
1444 */
1445int
1446hibernate_write_chunks(union hibernate_info *hib)
1447{
1448 paddr_t range_base, range_end, inaddr, temp_inaddr;
1449 size_t nblocks, out_remaining, used;
1450 struct hibernate_disk_chunk *chunks;
1451 vaddr_t hibernate_io_page = hib->piglet_va + PAGE_SIZE(1 << 12);
1452 daddr_t blkctr = 0;
1453 int i, rle, err;
7
'rle' declared without an initial value
1454 struct hibernate_zlib_state *hibernate_state;
1455
1456 hibernate_state =
1457 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
1458
1459 hib->chunk_ctr = 0;
1460
1461 /*
1462 * Map the utility VAs to the piglet. See the piglet map at the
1463 * top of this file for piglet layout information.
1464 */
1465 hibernate_copy_page = hib->piglet_va + 3 * PAGE_SIZE(1 << 12);
1466 hibernate_rle_page = hib->piglet_va + 28 * PAGE_SIZE(1 << 12);
1467
1468 chunks = (struct hibernate_disk_chunk *)(hib->piglet_va +
1469 HIBERNATE_CHUNK_SIZE0x400000);
1470
1471 /* Calculate the chunk regions */
1472 for (i = 0; i < hib->nranges; i++) {
8
Loop condition is true. Entering loop body
14
Loop condition is false. Execution continues on line 1491
1473 range_base = hib->ranges[i].base;
1474 range_end = hib->ranges[i].end;
1475
1476 inaddr = range_base;
1477
1478 while (inaddr
12.1
'inaddr' is >= 'range_end'
< range_end
) {
9
Assuming 'inaddr' is < 'range_end'
10
Loop condition is true. Entering loop body
13
Loop condition is false. Execution continues on line 1472
1479 chunks[hib->chunk_ctr].base = inaddr;
1480 if (inaddr + HIBERNATE_CHUNK_SIZE0x400000 < range_end)
11
Assuming the condition is false
12
Taking false branch
1481 chunks[hib->chunk_ctr].end = inaddr +
1482 HIBERNATE_CHUNK_SIZE0x400000;
1483 else
1484 chunks[hib->chunk_ctr].end = range_end;
1485
1486 inaddr += HIBERNATE_CHUNK_SIZE0x400000;
1487 hib->chunk_ctr ++;
1488 }
1489 }
1490
1491 uvm_pmr_dirty_everything();
1492 uvm_pmr_zero_everything();
1493
1494 /* Compress and write the chunks in the chunktable */
1495 for (i = 0; i < hib->chunk_ctr; i++) {
15
Loop condition is true. Entering loop body
1496 range_base = chunks[i].base;
1497 range_end = chunks[i].end;
1498
1499 chunks[i].offset = blkctr + hib->image_offset;
1500
1501 /* Reset zlib for deflate */
1502 if (hibernate_zlib_reset(hib, 1) != Z_OK0) {
16
Assuming the condition is false
17
Taking false branch
1503 DPRINTF("hibernate_zlib_reset failed for deflate\n");
1504 return (ENOMEM12);
1505 }
1506
1507 inaddr = range_base;
1508
1509 /*
1510 * For each range, loop through its phys mem region
1511 * and write out the chunks (the last chunk might be
1512 * smaller than the chunk size).
1513 */
1514 while (inaddr
17.1
'inaddr' is < 'range_end'
< range_end) {
18
Loop condition is true. Entering loop body
1515 out_remaining = PAGE_SIZE(1 << 12);
1516 while (out_remaining
18.1
'out_remaining' is > 0
> 0 && inaddr
18.2
'inaddr' is < 'range_end'
< range_end) {
19
Loop condition is true. Entering loop body
1517 /*
1518 * Adjust for regions that are not evenly
1519 * divisible by PAGE_SIZE or overflowed
1520 * pages from the previous iteration.
1521 */
1522 temp_inaddr = (inaddr & PAGE_MASK((1 << 12) - 1)) +
1523 hibernate_copy_page;
1524
1525 /* Deflate from temp_inaddr to IO page */
1526 if (inaddr
19.1
'inaddr' is not equal to 'range_end'
!= range_end) {
20
Taking true branch
1527 if (inaddr % PAGE_SIZE(1 << 12) == 0) {
21
Assuming the condition is false
22
Taking false branch
1528 rle = hibernate_write_rle(hib,
1529 inaddr,
1530 range_end,
1531 &blkctr,
1532 &out_remaining);
1533 }
1534
1535 if (rle == 0) {
23
The left operand of '==' is a garbage value
1536 pmap_kenter_pa(hibernate_temp_page,
1537 inaddr & PMAP_PA_MASK~((paddr_t)((1 << 12) - 1)),
1538 PROT_READ0x01);
1539
1540 bcopy((caddr_t)hibernate_temp_page,
1541 (caddr_t)hibernate_copy_page,
1542 PAGE_SIZE(1 << 12));
1543 inaddr += hibernate_deflate(hib,
1544 temp_inaddr,
1545 &out_remaining);
1546 } else {
1547 inaddr += rle * PAGE_SIZE(1 << 12);
1548 if (inaddr > range_end)
1549 inaddr = range_end;
1550 }
1551
1552 }
1553
1554 if (out_remaining == 0) {
1555 /* Filled up the page */
1556 nblocks = PAGE_SIZE(1 << 12) / DEV_BSIZE(1 << 9);
1557
1558 if ((err = hib->io_func(hib->dev,
1559 blkctr + hib->image_offset,
1560 (vaddr_t)hibernate_io_page,
1561 PAGE_SIZE(1 << 12), HIB_W1, hib->io_page))) {
1562 DPRINTF("hib write error %d\n",
1563 err);
1564 return (err);
1565 }
1566
1567 blkctr += nblocks;
1568 }
1569 }
1570 }
1571
1572 if (inaddr != range_end) {
1573 DPRINTF("deflate range ended prematurely\n");
1574 return (EINVAL22);
1575 }
1576
1577 /*
1578 * End of range. Round up to next secsize bytes
1579 * after finishing compress
1580 */
1581 if (out_remaining == 0)
1582 out_remaining = PAGE_SIZE(1 << 12);
1583
1584 /* Finish compress */
1585 hibernate_state->hib_stream.next_in = (unsigned char *)inaddr;
1586 hibernate_state->hib_stream.avail_in = 0;
1587 hibernate_state->hib_stream.next_out =
1588 (unsigned char *)hibernate_io_page +
1589 (PAGE_SIZE(1 << 12) - out_remaining);
1590
1591 /* We have an extra output page available for finalize */
1592 hibernate_state->hib_stream.avail_out =
1593 out_remaining + PAGE_SIZE(1 << 12);
1594
1595 if ((err = deflate(&hibernate_state->hib_stream, Z_FINISH4)) !=
1596 Z_STREAM_END1) {
1597 DPRINTF("deflate error in output stream: %d\n", err);
1598 return (err);
1599 }
1600
1601 out_remaining = hibernate_state->hib_stream.avail_out;
1602
1603 used = 2 * PAGE_SIZE(1 << 12) - out_remaining;
1604 nblocks = used / DEV_BSIZE(1 << 9);
1605
1606 /* Round up to next block if needed */
1607 if (used % DEV_BSIZE(1 << 9) != 0)
1608 nblocks ++;
1609
1610 /* Write final block(s) for this chunk */
1611 if ((err = hib->io_func(hib->dev, blkctr + hib->image_offset,
1612 (vaddr_t)hibernate_io_page, nblocks*DEV_BSIZE(1 << 9),
1613 HIB_W1, hib->io_page))) {
1614 DPRINTF("hib final write error %d\n", err);
1615 return (err);
1616 }
1617
1618 blkctr += nblocks;
1619
1620 chunks[i].compressed_size = (blkctr + hib->image_offset -
1621 chunks[i].offset) * DEV_BSIZE(1 << 9);
1622 }
1623
1624 hib->chunktable_offset = hib->image_offset + blkctr;
1625 return (0);
1626}
1627
1628/*
1629 * Reset the zlib stream state and allocate a new hiballoc area for either
1630 * inflate or deflate. This function is called once for each hibernate chunk.
1631 * Calling hiballoc_init multiple times is acceptable since the memory it is
1632 * provided is unmanaged memory (stolen). We use the memory provided to us
1633 * by the piglet allocated via the supplied hib.
1634 */
1635int
1636hibernate_zlib_reset(union hibernate_info *hib, int deflate)
1637{
1638 vaddr_t hibernate_zlib_start;
1639 size_t hibernate_zlib_size;
1640 char *pva = (char *)hib->piglet_va;
1641 struct hibernate_zlib_state *hibernate_state;
1642
1643 hibernate_state =
1644 (struct hibernate_zlib_state *)HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34);
1645
1646 if (!deflate)
1647 pva = (char *)((paddr_t)pva & (PIGLET_PAGE_MASK(((0x0000ff8000000000UL|0x0000007fc0000000UL)|0x000000003fe00000UL
))
));
1648
1649 /*
1650 * See piglet layout information at the start of this file for
1651 * information on the zlib page assignments.
1652 */
1653 hibernate_zlib_start = (vaddr_t)(pva + (30 * PAGE_SIZE(1 << 12)));
1654 hibernate_zlib_size = 80 * PAGE_SIZE(1 << 12);
1655
1656 memset((void *)hibernate_zlib_start, 0, hibernate_zlib_size)__builtin_memset(((void *)hibernate_zlib_start), (0), (hibernate_zlib_size
))
;
1657 memset(hibernate_state, 0, PAGE_SIZE)__builtin_memset((hibernate_state), (0), ((1 << 12)));
1658
1659 /* Set up stream structure */
1660 hibernate_state->hib_stream.zalloc = (alloc_func)hibernate_zlib_alloc;
1661 hibernate_state->hib_stream.zfree = (free_func)hibernate_zlib_free;
1662
1663 /* Initialize the hiballoc arena for zlib allocs/frees */
1664 hiballoc_init(&hibernate_state->hiballoc_arena,
1665 (caddr_t)hibernate_zlib_start, hibernate_zlib_size);
1666
1667 if (deflate) {
1668 return deflateInit(&hibernate_state->hib_stream,deflateInit_((&hibernate_state->hib_stream), (1), "1.2.11"
, (int)sizeof(z_stream))
1669 Z_BEST_SPEED)deflateInit_((&hibernate_state->hib_stream), (1), "1.2.11"
, (int)sizeof(z_stream))
;
1670 } else
1671 return inflateInit(&hibernate_state->hib_stream)inflateInit_((&hibernate_state->hib_stream), "1.2.11",
(int)sizeof(z_stream))
;
1672}
1673
1674/*
1675 * Reads the hibernated memory image from disk, whose location and
1676 * size are recorded in hib. Begin by reading the persisted
1677 * chunk table, which records the original chunk placement location
1678 * and compressed size for each. Next, allocate a pig region of
1679 * sufficient size to hold the compressed image. Next, read the
1680 * chunks into the pig area (calling hibernate_read_chunks to do this),
1681 * and finally, if all of the above succeeds, clear the hibernate signature.
1682 * The function will then return to hibernate_resume, which will proceed
1683 * to unpack the pig image to the correct place in memory.
1684 */
1685int
1686hibernate_read_image(union hibernate_info *hib)
1687{
1688 size_t compressed_size, disk_size, chunktable_size, pig_sz;
1689 paddr_t image_start, image_end, pig_start, pig_end;
1690 struct hibernate_disk_chunk *chunks;
1691 daddr_t blkctr;
1692 vaddr_t chunktable = (vaddr_t)NULL((void *)0);
1693 paddr_t piglet_chunktable = hib->piglet_pa +
1694 HIBERNATE_CHUNK_SIZE0x400000;
1695 int i, status;
1696
1697 status = 0;
1698 pmap_activate(curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1699
1700 /* Calculate total chunk table size in disk blocks */
1701 chunktable_size = HIBERNATE_CHUNK_TABLE_SIZE0x100000 / DEV_BSIZE(1 << 9);
1702
1703 blkctr = hib->chunktable_offset;
1704
1705 chunktable = (vaddr_t)km_alloc(HIBERNATE_CHUNK_TABLE_SIZE0x100000, &kv_any,
1706 &kp_none, &kd_nowait);
1707
1708 if (!chunktable)
1709 return (1);
1710
1711 /* Map chunktable pages */
1712 for (i = 0; i < HIBERNATE_CHUNK_TABLE_SIZE0x100000; i += PAGE_SIZE(1 << 12))
1713 pmap_kenter_pa(chunktable + i, piglet_chunktable + i,
1714 PROT_READ0x01 | PROT_WRITE0x02);
1715 pmap_update(pmap_kernel());
1716
1717 /* Read the chunktable from disk into the piglet chunktable */
1718 for (i = 0; i < HIBERNATE_CHUNK_TABLE_SIZE0x100000;
1719 i += MAXPHYS(64 * 1024), blkctr += MAXPHYS(64 * 1024)/DEV_BSIZE(1 << 9))
1720 hibernate_block_io(hib, blkctr, MAXPHYS(64 * 1024),
1721 chunktable + i, 0);
1722
1723 blkctr = hib->image_offset;
1724 compressed_size = 0;
1725
1726 chunks = (struct hibernate_disk_chunk *)chunktable;
1727
1728 for (i = 0; i < hib->chunk_ctr; i++)
1729 compressed_size += chunks[i].compressed_size;
1730
1731 disk_size = compressed_size;
1732
1733 printf("unhibernating @ block %lld length %luMB\n",
1734 hib->sig_offset - chunktable_size,
1735 compressed_size / (1024 * 1024));
1736
1737 /* Allocate the pig area */
1738 pig_sz = compressed_size + HIBERNATE_CHUNK_SIZE0x400000;
1739 if (uvm_pmr_alloc_pig(&pig_start, pig_sz, hib->piglet_pa) == ENOMEM12) {
1740 status = 1;
1741 goto unmap;
1742 }
1743
1744 pig_end = pig_start + pig_sz;
1745
1746 /* Calculate image extents. Pig image must end on a chunk boundary. */
1747 image_end = pig_end & ~(HIBERNATE_CHUNK_SIZE0x400000 - 1);
1748 image_start = image_end - disk_size;
1749
1750 hibernate_read_chunks(hib, image_start, image_end, disk_size,
1751 chunks);
1752
1753 /* Prepare the resume time pmap/page table */
1754 hibernate_populate_resume_pt(hib, image_start, image_end);
1755
1756unmap:
1757 /* Unmap chunktable pages */
1758 pmap_kremove(chunktable, HIBERNATE_CHUNK_TABLE_SIZE0x100000);
1759 pmap_update(pmap_kernel());
1760
1761 return (status);
1762}
1763
1764/*
1765 * Read the hibernated memory chunks from disk (chunk information at this
1766 * point is stored in the piglet) into the pig area specified by
1767 * [pig_start .. pig_end]. Order the chunks so that the final chunk is the
1768 * only chunk with overlap possibilities.
1769 */
1770int
1771hibernate_read_chunks(union hibernate_info *hib, paddr_t pig_start,
1772 paddr_t pig_end, size_t image_compr_size,
1773 struct hibernate_disk_chunk *chunks)
1774{
1775 paddr_t img_cur, piglet_base;
1776 daddr_t blkctr;
1777 size_t processed, compressed_size, read_size;
1778 int nchunks, nfchunks, num_io_pages;
1779 vaddr_t tempva, hibernate_fchunk_area;
1780 short *fchunks, i, j;
1781
1782 tempva = (vaddr_t)NULL((void *)0);
1783 hibernate_fchunk_area = (vaddr_t)NULL((void *)0);
1784 nfchunks = 0;
1785 piglet_base = hib->piglet_pa;
1786 global_pig_start = pig_start;
1787
1788 /*
1789 * These mappings go into the resuming kernel's page table, and are
1790 * used only during image read. They disappear from existence
1791 * when the suspended kernel is unpacked on top of us.
1792 */
1793 tempva = (vaddr_t)km_alloc(MAXPHYS(64 * 1024) + PAGE_SIZE(1 << 12), &kv_any, &kp_none,
1794 &kd_nowait);
1795 if (!tempva)
1796 return (1);
1797 hibernate_fchunk_area = (vaddr_t)km_alloc(24 * PAGE_SIZE(1 << 12), &kv_any,
1798 &kp_none, &kd_nowait);
1799 if (!hibernate_fchunk_area)
1800 return (1);
1801
1802 /* Final output chunk ordering VA */
1803 fchunks = (short *)hibernate_fchunk_area;
1804
1805 /* Map the chunk ordering region */
1806 for(i = 0; i < 24 ; i++)
1807 pmap_kenter_pa(hibernate_fchunk_area + (i * PAGE_SIZE(1 << 12)),
1808 piglet_base + ((4 + i) * PAGE_SIZE(1 << 12)),
1809 PROT_READ0x01 | PROT_WRITE0x02);
1810 pmap_update(pmap_kernel());
1811
1812 nchunks = hib->chunk_ctr;
1813
1814 /* Initially start all chunks as unplaced */
1815 for (i = 0; i < nchunks; i++)
1816 chunks[i].flags = 0;
1817
1818 /*
1819 * Search the list for chunks that are outside the pig area. These
1820 * can be placed first in the final output list.
1821 */
1822 for (i = 0; i < nchunks; i++) {
1823 if (chunks[i].end <= pig_start || chunks[i].base >= pig_end) {
1824 fchunks[nfchunks] = i;
1825 nfchunks++;
1826 chunks[i].flags |= HIBERNATE_CHUNK_PLACED4;
1827 }
1828 }
1829
1830 /*
1831 * Walk the ordering, place the chunks in ascending memory order.
1832 */
1833 for (i = 0; i < nchunks; i++) {
1834 if (chunks[i].flags != HIBERNATE_CHUNK_PLACED4) {
1835 fchunks[nfchunks] = i;
1836 nfchunks++;
1837 chunks[i].flags = HIBERNATE_CHUNK_PLACED4;
1838 }
1839 }
1840
1841 img_cur = pig_start;
1842
1843 for (i = 0; i < nfchunks; i++) {
1844 blkctr = chunks[fchunks[i]].offset;
1845 processed = 0;
1846 compressed_size = chunks[fchunks[i]].compressed_size;
1847
1848 while (processed < compressed_size) {
1849 if (compressed_size - processed >= MAXPHYS(64 * 1024))
1850 read_size = MAXPHYS(64 * 1024);
1851 else
1852 read_size = compressed_size - processed;
1853
1854 /*
1855 * We're reading read_size bytes, offset from the
1856 * start of a page by img_cur % PAGE_SIZE, so the
1857 * end will be read_size + (img_cur % PAGE_SIZE)
1858 * from the start of the first page. Round that
1859 * up to the next page size.
1860 */
1861 num_io_pages = (read_size + (img_cur % PAGE_SIZE(1 << 12))
1862 + PAGE_SIZE(1 << 12) - 1) / PAGE_SIZE(1 << 12);
1863
1864 KASSERT(num_io_pages <= MAXPHYS/PAGE_SIZE + 1)((num_io_pages <= (64 * 1024)/(1 << 12) + 1) ? (void
)0 : __assert("diagnostic ", "/usr/src/sys/kern/subr_hibernate.c"
, 1864, "num_io_pages <= MAXPHYS/PAGE_SIZE + 1"))
;
1865
1866 /* Map pages for this read */
1867 for (j = 0; j < num_io_pages; j ++)
1868 pmap_kenter_pa(tempva + j * PAGE_SIZE(1 << 12),
1869 img_cur + j * PAGE_SIZE(1 << 12),
1870 PROT_READ0x01 | PROT_WRITE0x02);
1871
1872 pmap_update(pmap_kernel());
1873
1874 hibernate_block_io(hib, blkctr, read_size,
1875 tempva + (img_cur & PAGE_MASK((1 << 12) - 1)), 0);
1876
1877 blkctr += (read_size / DEV_BSIZE(1 << 9));
1878
1879 pmap_kremove(tempva, num_io_pages * PAGE_SIZE(1 << 12));
1880 pmap_update(pmap_kernel());
1881
1882 processed += read_size;
1883 img_cur += read_size;
1884 }
1885 }
1886
1887 pmap_kremove(hibernate_fchunk_area, 24 * PAGE_SIZE(1 << 12));
1888 pmap_update(pmap_kernel());
1889
1890 return (0);
1891}
1892
1893/*
1894 * Hibernating a machine comprises the following operations:
1895 * 1. Calculating this machine's hibernate_info information
1896 * 2. Allocating a piglet and saving the piglet's physaddr
1897 * 3. Calculating the memory chunks
1898 * 4. Writing the compressed chunks to disk
1899 * 5. Writing the chunk table
1900 * 6. Writing the signature block (hibernate_info)
1901 *
1902 * On most architectures, the function calling hibernate_suspend would
1903 * then power off the machine using some MD-specific implementation.
1904 */
1905int
1906hibernate_suspend(void)
1907{
1908 union hibernate_info hib;
1909 u_long start, end;
1910
1911 /*
1912 * Calculate memory ranges, swap offsets, etc.
1913 * This also allocates a piglet whose physaddr is stored in
1914 * hib->piglet_pa and vaddr stored in hib->piglet_va
1915 */
1916 if (get_hibernate_info(&hib, 1)) {
1
Taking false branch
1917 DPRINTF("failed to obtain hibernate info\n");
1918 return (1);
1919 }
1920
1921 /* Find a page-addressed region in swap [start,end] */
1922 if (uvm_hibswap(hib.dev, &start, &end)) {
2
Assuming the condition is false
3
Taking false branch
1923 printf("hibernate: cannot find any swap\n");
1924 return (1);
1925 }
1926
1927 if (end - start < 1000) {
4
Assuming the condition is false
5
Taking false branch
1928 printf("hibernate: insufficient swap (%lu is too small)\n",
1929 end - start);
1930 return (1);
1931 }
1932
1933 /* Calculate block offsets in swap */
1934 hib.image_offset = ctod(start)((start) << (12 - 9));
1935
1936 DPRINTF("hibernate @ block %lld max-length %lu blocks\n",
1937 hib.image_offset, ctod(end) - ctod(start));
1938
1939 pmap_activate(curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1940 DPRINTF("hibernate: writing chunks\n");
1941 if (hibernate_write_chunks(&hib)) {
6
Calling 'hibernate_write_chunks'
1942 DPRINTF("hibernate_write_chunks failed\n");
1943 return (1);
1944 }
1945
1946 DPRINTF("hibernate: writing chunktable\n");
1947 if (hibernate_write_chunktable(&hib)) {
1948 DPRINTF("hibernate_write_chunktable failed\n");
1949 return (1);
1950 }
1951
1952 DPRINTF("hibernate: writing signature\n");
1953 if (hibernate_write_signature(&hib)) {
1954 DPRINTF("hibernate_write_signature failed\n");
1955 return (1);
1956 }
1957
1958 /* Allow the disk to settle */
1959 delay(500000)(*delay_func)(500000);
1960
1961 /*
1962 * Give the device-specific I/O function a notification that we're
1963 * done, and that it can clean up or shutdown as needed.
1964 */
1965 hib.io_func(hib.dev, 0, (vaddr_t)NULL((void *)0), 0, HIB_DONE-2, hib.io_page);
1966 return (0);
1967}
1968
1969int
1970hibernate_alloc(void)
1971{
1972 KASSERT(global_piglet_va == 0)((global_piglet_va == 0) ? (void)0 : __assert("diagnostic ", "/usr/src/sys/kern/subr_hibernate.c"
, 1972, "global_piglet_va == 0"))
;
1973 KASSERT(hibernate_temp_page == 0)((hibernate_temp_page == 0) ? (void)0 : __assert("diagnostic "
, "/usr/src/sys/kern/subr_hibernate.c", 1973, "hibernate_temp_page == 0"
))
;
1974
1975 pmap_activate(curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
1976 pmap_kenter_pa(HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34), HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34),
1977 PROT_READ0x01 | PROT_WRITE0x02);
1978
1979 /* Allocate a piglet, store its addresses in the supplied globals */
1980 if (uvm_pmr_alloc_piglet(&global_piglet_va, &global_piglet_pa,
1981 HIBERNATE_CHUNK_SIZE0x400000 * 4, HIBERNATE_CHUNK_SIZE0x400000))
1982 goto unmap;
1983
1984 /*
1985 * Allocate VA for the temp page.
1986 *
1987 * This will become part of the suspended kernel and will
1988 * be freed in hibernate_free, upon resume (or hibernate
1989 * failure)
1990 */
1991 hibernate_temp_page = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_any,
1992 &kp_none, &kd_nowait);
1993 if (!hibernate_temp_page) {
1994 uvm_pmr_free_piglet(global_piglet_va,
1995 4 * HIBERNATE_CHUNK_SIZE0x400000);
1996 global_piglet_va = 0;
1997 goto unmap;
1998 }
1999 return (0);
2000unmap:
2001 pmap_kremove(HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34), PAGE_SIZE(1 << 12));
2002 pmap_update(pmap_kernel());
2003 return (ENOMEM12);
2004}
2005
2006/*
2007 * Free items allocated by hibernate_alloc()
2008 */
2009void
2010hibernate_free(void)
2011{
2012 pmap_activate(curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc
);
2013
2014 if (global_piglet_va)
2015 uvm_pmr_free_piglet(global_piglet_va,
2016 4 * HIBERNATE_CHUNK_SIZE0x400000);
2017
2018 if (hibernate_temp_page) {
2019 pmap_kremove(hibernate_temp_page, PAGE_SIZE(1 << 12));
2020 km_free((void *)hibernate_temp_page, PAGE_SIZE(1 << 12),
2021 &kv_any, &kp_none);
2022 }
2023
2024 global_piglet_va = 0;
2025 hibernate_temp_page = 0;
2026 pmap_kremove(HIBERNATE_HIBALLOC_PAGE((1 << 12) * 34), PAGE_SIZE(1 << 12));
2027 pmap_update(pmap_kernel());
2028}