Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
torvalds
GitHub Repository: torvalds/linux
Path: blob/master/arch/powerpc/mm/ptdump/hashpagetable.c
26481 views
1
// SPDX-License-Identifier: GPL-2.0-only
2
/*
3
* Copyright 2016, Rashmica Gupta, IBM Corp.
4
*
5
* This traverses the kernel virtual memory and dumps the pages that are in
6
* the hash pagetable, along with their flags to
7
* /sys/kernel/debug/kernel_hash_pagetable.
8
*
9
* If radix is enabled then there is no hash page table and so no debugfs file
10
* is generated.
11
*/
12
#include <linux/debugfs.h>
13
#include <linux/fs.h>
14
#include <linux/io.h>
15
#include <linux/mm.h>
16
#include <linux/sched.h>
17
#include <linux/seq_file.h>
18
#include <linux/const.h>
19
#include <asm/page.h>
20
#include <asm/plpar_wrappers.h>
21
#include <linux/memblock.h>
22
#include <asm/firmware.h>
23
#include <asm/pgalloc.h>
24
25
struct pg_state {
26
struct seq_file *seq;
27
const struct addr_marker *marker;
28
unsigned long start_address;
29
unsigned int level;
30
u64 current_flags;
31
};
32
33
struct addr_marker {
34
unsigned long start_address;
35
const char *name;
36
};
37
38
static struct addr_marker address_markers[] = {
39
{ 0, "Start of kernel VM" },
40
{ 0, "vmalloc() Area" },
41
{ 0, "vmalloc() End" },
42
{ 0, "isa I/O start" },
43
{ 0, "isa I/O end" },
44
{ 0, "phb I/O start" },
45
{ 0, "phb I/O end" },
46
{ 0, "I/O remap start" },
47
{ 0, "I/O remap end" },
48
{ 0, "vmemmap start" },
49
{ -1, NULL },
50
};
51
52
struct flag_info {
53
u64 mask;
54
u64 val;
55
const char *set;
56
const char *clear;
57
bool is_val;
58
int shift;
59
};
60
61
static const struct flag_info v_flag_array[] = {
62
{
63
.mask = SLB_VSID_B,
64
.val = SLB_VSID_B_256M,
65
.set = "ssize: 256M",
66
.clear = "ssize: 1T ",
67
}, {
68
.mask = HPTE_V_SECONDARY,
69
.val = HPTE_V_SECONDARY,
70
.set = "secondary",
71
.clear = "primary ",
72
}, {
73
.mask = HPTE_V_VALID,
74
.val = HPTE_V_VALID,
75
.set = "valid ",
76
.clear = "invalid",
77
}, {
78
.mask = HPTE_V_BOLTED,
79
.val = HPTE_V_BOLTED,
80
.set = "bolted",
81
.clear = "",
82
}
83
};
84
85
static const struct flag_info r_flag_array[] = {
86
{
87
.mask = HPTE_R_PP0 | HPTE_R_PP,
88
.val = PP_RWXX,
89
.set = "prot:RW--",
90
}, {
91
.mask = HPTE_R_PP0 | HPTE_R_PP,
92
.val = PP_RWRX,
93
.set = "prot:RWR-",
94
}, {
95
.mask = HPTE_R_PP0 | HPTE_R_PP,
96
.val = PP_RWRW,
97
.set = "prot:RWRW",
98
}, {
99
.mask = HPTE_R_PP0 | HPTE_R_PP,
100
.val = PP_RXRX,
101
.set = "prot:R-R-",
102
}, {
103
.mask = HPTE_R_PP0 | HPTE_R_PP,
104
.val = PP_RXXX,
105
.set = "prot:R---",
106
}, {
107
.mask = HPTE_R_KEY_HI | HPTE_R_KEY_LO,
108
.val = HPTE_R_KEY_HI | HPTE_R_KEY_LO,
109
.set = "key",
110
.clear = "",
111
.is_val = true,
112
}, {
113
.mask = HPTE_R_R,
114
.val = HPTE_R_R,
115
.set = "ref",
116
.clear = " ",
117
}, {
118
.mask = HPTE_R_C,
119
.val = HPTE_R_C,
120
.set = "changed",
121
.clear = " ",
122
}, {
123
.mask = HPTE_R_N,
124
.val = HPTE_R_N,
125
.set = "no execute",
126
}, {
127
.mask = HPTE_R_WIMG,
128
.val = HPTE_R_W,
129
.set = "writethru",
130
}, {
131
.mask = HPTE_R_WIMG,
132
.val = HPTE_R_I,
133
.set = "no cache",
134
}, {
135
.mask = HPTE_R_WIMG,
136
.val = HPTE_R_G,
137
.set = "guarded",
138
}
139
};
140
141
static int calculate_pagesize(struct pg_state *st, int ps, char s[])
142
{
143
static const char units[] = "BKMGTPE";
144
const char *unit = units;
145
146
while (ps > 9 && unit[1]) {
147
ps -= 10;
148
unit++;
149
}
150
seq_printf(st->seq, " %s_ps: %i%c\t", s, 1<<ps, *unit);
151
return ps;
152
}
153
154
static void dump_flag_info(struct pg_state *st, const struct flag_info
155
*flag, u64 pte, int num)
156
{
157
unsigned int i;
158
159
for (i = 0; i < num; i++, flag++) {
160
const char *s = NULL;
161
u64 val;
162
163
/* flag not defined so don't check it */
164
if (flag->mask == 0)
165
continue;
166
/* Some 'flags' are actually values */
167
if (flag->is_val) {
168
val = pte & flag->val;
169
if (flag->shift)
170
val = val >> flag->shift;
171
seq_printf(st->seq, " %s:%llx", flag->set, val);
172
} else {
173
if ((pte & flag->mask) == flag->val)
174
s = flag->set;
175
else
176
s = flag->clear;
177
if (s)
178
seq_printf(st->seq, " %s", s);
179
}
180
}
181
}
182
183
static void dump_hpte_info(struct pg_state *st, unsigned long ea, u64 v, u64 r,
184
unsigned long rpn, int bps, int aps, unsigned long lp)
185
{
186
int aps_index;
187
188
while (ea >= st->marker[1].start_address) {
189
st->marker++;
190
seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
191
}
192
seq_printf(st->seq, "0x%lx:\t", ea);
193
seq_printf(st->seq, "AVPN:%llx\t", HPTE_V_AVPN_VAL(v));
194
dump_flag_info(st, v_flag_array, v, ARRAY_SIZE(v_flag_array));
195
seq_printf(st->seq, " rpn: %lx\t", rpn);
196
dump_flag_info(st, r_flag_array, r, ARRAY_SIZE(r_flag_array));
197
198
calculate_pagesize(st, bps, "base");
199
aps_index = calculate_pagesize(st, aps, "actual");
200
if (aps_index != 2)
201
seq_printf(st->seq, "LP enc: %lx", lp);
202
seq_putc(st->seq, '\n');
203
}
204
205
206
static int native_find(unsigned long ea, int psize, bool primary, u64 *v, u64
207
*r)
208
{
209
struct hash_pte *hptep;
210
unsigned long hash, vsid, vpn, hpte_group, want_v, hpte_v;
211
int i, ssize = mmu_kernel_ssize;
212
unsigned long shift = mmu_psize_defs[psize].shift;
213
214
/* calculate hash */
215
vsid = get_kernel_vsid(ea, ssize);
216
vpn = hpt_vpn(ea, vsid, ssize);
217
hash = hpt_hash(vpn, shift, ssize);
218
want_v = hpte_encode_avpn(vpn, psize, ssize);
219
220
/* to check in the secondary hash table, we invert the hash */
221
if (!primary)
222
hash = ~hash;
223
hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
224
for (i = 0; i < HPTES_PER_GROUP; i++) {
225
hptep = htab_address + hpte_group;
226
hpte_v = be64_to_cpu(hptep->v);
227
228
if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
229
/* HPTE matches */
230
*v = be64_to_cpu(hptep->v);
231
*r = be64_to_cpu(hptep->r);
232
return 0;
233
}
234
++hpte_group;
235
}
236
return -1;
237
}
238
239
static int pseries_find(unsigned long ea, int psize, bool primary, u64 *v, u64 *r)
240
{
241
struct {
242
unsigned long v;
243
unsigned long r;
244
} ptes[4];
245
unsigned long vsid, vpn, hash, hpte_group, want_v;
246
int i, j, ssize = mmu_kernel_ssize;
247
long lpar_rc = 0;
248
unsigned long shift = mmu_psize_defs[psize].shift;
249
250
/* calculate hash */
251
vsid = get_kernel_vsid(ea, ssize);
252
vpn = hpt_vpn(ea, vsid, ssize);
253
hash = hpt_hash(vpn, shift, ssize);
254
want_v = hpte_encode_avpn(vpn, psize, ssize);
255
256
/* to check in the secondary hash table, we invert the hash */
257
if (!primary)
258
hash = ~hash;
259
hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
260
/* see if we can find an entry in the hpte with this hash */
261
for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
262
lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
263
264
if (lpar_rc)
265
continue;
266
for (j = 0; j < 4; j++) {
267
if (HPTE_V_COMPARE(ptes[j].v, want_v) &&
268
(ptes[j].v & HPTE_V_VALID)) {
269
/* HPTE matches */
270
*v = ptes[j].v;
271
*r = ptes[j].r;
272
return 0;
273
}
274
}
275
}
276
return -1;
277
}
278
279
static void decode_r(int bps, unsigned long r, unsigned long *rpn, int *aps,
280
unsigned long *lp_bits)
281
{
282
struct mmu_psize_def entry;
283
unsigned long arpn, mask, lp;
284
int penc = -2, idx = 0, shift;
285
286
/*.
287
* The LP field has 8 bits. Depending on the actual page size, some of
288
* these bits are concatenated with the APRN to get the RPN. The rest
289
* of the bits in the LP field is the LP value and is an encoding for
290
* the base page size and the actual page size.
291
*
292
* - find the mmu entry for our base page size
293
* - go through all page encodings and use the associated mask to
294
* find an encoding that matches our encoding in the LP field.
295
*/
296
arpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
297
lp = arpn & 0xff;
298
299
entry = mmu_psize_defs[bps];
300
while (idx < MMU_PAGE_COUNT) {
301
penc = entry.penc[idx];
302
if ((penc != -1) && (mmu_psize_defs[idx].shift)) {
303
shift = mmu_psize_defs[idx].shift - HPTE_R_RPN_SHIFT;
304
mask = (0x1 << (shift)) - 1;
305
if ((lp & mask) == penc) {
306
*aps = mmu_psize_to_shift(idx);
307
*lp_bits = lp & mask;
308
*rpn = arpn >> shift;
309
return;
310
}
311
}
312
idx++;
313
}
314
}
315
316
static int base_hpte_find(unsigned long ea, int psize, bool primary, u64 *v,
317
u64 *r)
318
{
319
if (IS_ENABLED(CONFIG_PPC_PSERIES) && firmware_has_feature(FW_FEATURE_LPAR))
320
return pseries_find(ea, psize, primary, v, r);
321
322
return native_find(ea, psize, primary, v, r);
323
}
324
325
static unsigned long hpte_find(struct pg_state *st, unsigned long ea, int psize)
326
{
327
unsigned long slot;
328
u64 v = 0, r = 0;
329
unsigned long rpn, lp_bits;
330
int base_psize = 0, actual_psize = 0;
331
332
if (ea < PAGE_OFFSET)
333
return -1;
334
335
/* Look in primary table */
336
slot = base_hpte_find(ea, psize, true, &v, &r);
337
338
/* Look in secondary table */
339
if (slot == -1)
340
slot = base_hpte_find(ea, psize, false, &v, &r);
341
342
/* No entry found */
343
if (slot == -1)
344
return -1;
345
346
/*
347
* We found an entry in the hash page table:
348
* - check that this has the same base page
349
* - find the actual page size
350
* - find the RPN
351
*/
352
base_psize = mmu_psize_to_shift(psize);
353
354
if ((v & HPTE_V_LARGE) == HPTE_V_LARGE) {
355
decode_r(psize, r, &rpn, &actual_psize, &lp_bits);
356
} else {
357
/* 4K actual page size */
358
actual_psize = 12;
359
rpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
360
/* In this case there are no LP bits */
361
lp_bits = -1;
362
}
363
/*
364
* We didn't find a matching encoding, so the PTE we found isn't for
365
* this address.
366
*/
367
if (actual_psize == -1)
368
return -1;
369
370
dump_hpte_info(st, ea, v, r, rpn, base_psize, actual_psize, lp_bits);
371
return 0;
372
}
373
374
static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
375
{
376
pte_t *pte = pte_offset_kernel(pmd, 0);
377
unsigned long addr, pteval, psize;
378
int i, status;
379
380
for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
381
addr = start + i * PAGE_SIZE;
382
pteval = pte_val(*pte);
383
384
if (addr < VMALLOC_END)
385
psize = mmu_vmalloc_psize;
386
else
387
psize = mmu_io_psize;
388
389
/* check for secret 4K mappings */
390
if (IS_ENABLED(CONFIG_PPC_64K_PAGES) &&
391
((pteval & H_PAGE_COMBO) == H_PAGE_COMBO ||
392
(pteval & H_PAGE_4K_PFN) == H_PAGE_4K_PFN))
393
psize = mmu_io_psize;
394
395
/* check for hashpte */
396
status = hpte_find(st, addr, psize);
397
398
if (((pteval & H_PAGE_HASHPTE) != H_PAGE_HASHPTE)
399
&& (status != -1)) {
400
/* found a hpte that is not in the linux page tables */
401
seq_printf(st->seq, "page probably bolted before linux"
402
" pagetables were set: addr:%lx, pteval:%lx\n",
403
addr, pteval);
404
}
405
}
406
}
407
408
static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
409
{
410
pmd_t *pmd = pmd_offset(pud, 0);
411
unsigned long addr;
412
unsigned int i;
413
414
for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
415
addr = start + i * PMD_SIZE;
416
if (!pmd_none(*pmd))
417
/* pmd exists */
418
walk_pte(st, pmd, addr);
419
}
420
}
421
422
static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
423
{
424
pud_t *pud = pud_offset(p4d, 0);
425
unsigned long addr;
426
unsigned int i;
427
428
for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
429
addr = start + i * PUD_SIZE;
430
if (!pud_none(*pud))
431
/* pud exists */
432
walk_pmd(st, pud, addr);
433
}
434
}
435
436
static void walk_p4d(struct pg_state *st, pgd_t *pgd, unsigned long start)
437
{
438
p4d_t *p4d = p4d_offset(pgd, 0);
439
unsigned long addr;
440
unsigned int i;
441
442
for (i = 0; i < PTRS_PER_P4D; i++, p4d++) {
443
addr = start + i * P4D_SIZE;
444
if (!p4d_none(*p4d))
445
/* p4d exists */
446
walk_pud(st, p4d, addr);
447
}
448
}
449
450
static void walk_pagetables(struct pg_state *st)
451
{
452
pgd_t *pgd = pgd_offset_k(0UL);
453
unsigned int i;
454
unsigned long addr;
455
456
/*
457
* Traverse the linux pagetable structure and dump pages that are in
458
* the hash pagetable.
459
*/
460
for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
461
addr = KERN_VIRT_START + i * PGDIR_SIZE;
462
if (!pgd_none(*pgd))
463
/* pgd exists */
464
walk_p4d(st, pgd, addr);
465
}
466
}
467
468
469
static void walk_linearmapping(struct pg_state *st)
470
{
471
unsigned long addr;
472
473
/*
474
* Traverse the linear mapping section of virtual memory and dump pages
475
* that are in the hash pagetable.
476
*/
477
unsigned long psize = 1 << mmu_psize_defs[mmu_linear_psize].shift;
478
479
for (addr = PAGE_OFFSET; addr < PAGE_OFFSET +
480
memblock_end_of_DRAM(); addr += psize)
481
hpte_find(st, addr, mmu_linear_psize);
482
}
483
484
static void walk_vmemmap(struct pg_state *st)
485
{
486
struct vmemmap_backing *ptr = vmemmap_list;
487
488
if (!IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
489
return;
490
/*
491
* Traverse the vmemmaped memory and dump pages that are in the hash
492
* pagetable.
493
*/
494
while (ptr) {
495
hpte_find(st, ptr->virt_addr, mmu_vmemmap_psize);
496
ptr = ptr->list;
497
}
498
seq_puts(st->seq, "---[ vmemmap end ]---\n");
499
}
500
501
static void populate_markers(void)
502
{
503
address_markers[0].start_address = PAGE_OFFSET;
504
address_markers[1].start_address = VMALLOC_START;
505
address_markers[2].start_address = VMALLOC_END;
506
address_markers[3].start_address = ISA_IO_BASE;
507
address_markers[4].start_address = ISA_IO_END;
508
address_markers[5].start_address = PHB_IO_BASE;
509
address_markers[6].start_address = PHB_IO_END;
510
address_markers[7].start_address = IOREMAP_BASE;
511
address_markers[8].start_address = IOREMAP_END;
512
address_markers[9].start_address = H_VMEMMAP_START;
513
}
514
515
static int ptdump_show(struct seq_file *m, void *v)
516
{
517
struct pg_state st = {
518
.seq = m,
519
.start_address = PAGE_OFFSET,
520
.marker = address_markers,
521
};
522
/*
523
* Traverse the 0xc, 0xd and 0xf areas of the kernel virtual memory and
524
* dump pages that are in the hash pagetable.
525
*/
526
walk_linearmapping(&st);
527
walk_pagetables(&st);
528
walk_vmemmap(&st);
529
return 0;
530
}
531
532
DEFINE_SHOW_ATTRIBUTE(ptdump);
533
534
static int ptdump_init(void)
535
{
536
if (!radix_enabled()) {
537
populate_markers();
538
debugfs_create_file("kernel_hash_pagetable", 0400, NULL, NULL,
539
&ptdump_fops);
540
}
541
return 0;
542
}
543
device_initcall(ptdump_init);
544
545