Loading...
Note: File does not exist in v6.8.
1/*
2 * Copyright 2016, Rashmica Gupta, IBM Corp.
3 *
4 * This traverses the kernel virtual memory and dumps the pages that are in
5 * the hash pagetable, along with their flags to
6 * /sys/kernel/debug/kernel_hash_pagetable.
7 *
8 * If radix is enabled then there is no hash page table and so no debugfs file
9 * is generated.
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; version 2
14 * of the License.
15 */
16#include <linux/debugfs.h>
17#include <linux/fs.h>
18#include <linux/io.h>
19#include <linux/mm.h>
20#include <linux/sched.h>
21#include <linux/seq_file.h>
22#include <asm/fixmap.h>
23#include <asm/pgtable.h>
24#include <linux/const.h>
25#include <asm/page.h>
26#include <asm/pgalloc.h>
27#include <asm/plpar_wrappers.h>
28#include <linux/memblock.h>
29#include <asm/firmware.h>
30
31struct pg_state {
32 struct seq_file *seq;
33 const struct addr_marker *marker;
34 unsigned long start_address;
35 unsigned int level;
36 u64 current_flags;
37};
38
39struct addr_marker {
40 unsigned long start_address;
41 const char *name;
42};
43
44static struct addr_marker address_markers[] = {
45 { 0, "Start of kernel VM" },
46 { 0, "vmalloc() Area" },
47 { 0, "vmalloc() End" },
48 { 0, "isa I/O start" },
49 { 0, "isa I/O end" },
50 { 0, "phb I/O start" },
51 { 0, "phb I/O end" },
52 { 0, "I/O remap start" },
53 { 0, "I/O remap end" },
54 { 0, "vmemmap start" },
55 { -1, NULL },
56};
57
58struct flag_info {
59 u64 mask;
60 u64 val;
61 const char *set;
62 const char *clear;
63 bool is_val;
64 int shift;
65};
66
67static const struct flag_info v_flag_array[] = {
68 {
69 .mask = SLB_VSID_B,
70 .val = SLB_VSID_B_256M,
71 .set = "ssize: 256M",
72 .clear = "ssize: 1T ",
73 }, {
74 .mask = HPTE_V_SECONDARY,
75 .val = HPTE_V_SECONDARY,
76 .set = "secondary",
77 .clear = "primary ",
78 }, {
79 .mask = HPTE_V_VALID,
80 .val = HPTE_V_VALID,
81 .set = "valid ",
82 .clear = "invalid",
83 }, {
84 .mask = HPTE_V_BOLTED,
85 .val = HPTE_V_BOLTED,
86 .set = "bolted",
87 .clear = "",
88 }
89};
90
91static const struct flag_info r_flag_array[] = {
92 {
93 .mask = HPTE_R_PP0 | HPTE_R_PP,
94 .val = PP_RWXX,
95 .set = "prot:RW--",
96 }, {
97 .mask = HPTE_R_PP0 | HPTE_R_PP,
98 .val = PP_RWRX,
99 .set = "prot:RWR-",
100 }, {
101 .mask = HPTE_R_PP0 | HPTE_R_PP,
102 .val = PP_RWRW,
103 .set = "prot:RWRW",
104 }, {
105 .mask = HPTE_R_PP0 | HPTE_R_PP,
106 .val = PP_RXRX,
107 .set = "prot:R-R-",
108 }, {
109 .mask = HPTE_R_PP0 | HPTE_R_PP,
110 .val = PP_RXXX,
111 .set = "prot:R---",
112 }, {
113 .mask = HPTE_R_KEY_HI | HPTE_R_KEY_LO,
114 .val = HPTE_R_KEY_HI | HPTE_R_KEY_LO,
115 .set = "key",
116 .clear = "",
117 .is_val = true,
118 }, {
119 .mask = HPTE_R_R,
120 .val = HPTE_R_R,
121 .set = "ref",
122 .clear = " ",
123 }, {
124 .mask = HPTE_R_C,
125 .val = HPTE_R_C,
126 .set = "changed",
127 .clear = " ",
128 }, {
129 .mask = HPTE_R_N,
130 .val = HPTE_R_N,
131 .set = "no execute",
132 }, {
133 .mask = HPTE_R_WIMG,
134 .val = HPTE_R_W,
135 .set = "writethru",
136 }, {
137 .mask = HPTE_R_WIMG,
138 .val = HPTE_R_I,
139 .set = "no cache",
140 }, {
141 .mask = HPTE_R_WIMG,
142 .val = HPTE_R_G,
143 .set = "guarded",
144 }
145};
146
147static int calculate_pagesize(struct pg_state *st, int ps, char s[])
148{
149 static const char units[] = "BKMGTPE";
150 const char *unit = units;
151
152 while (ps > 9 && unit[1]) {
153 ps -= 10;
154 unit++;
155 }
156 seq_printf(st->seq, " %s_ps: %i%c\t", s, 1<<ps, *unit);
157 return ps;
158}
159
160static void dump_flag_info(struct pg_state *st, const struct flag_info
161 *flag, u64 pte, int num)
162{
163 unsigned int i;
164
165 for (i = 0; i < num; i++, flag++) {
166 const char *s = NULL;
167 u64 val;
168
169 /* flag not defined so don't check it */
170 if (flag->mask == 0)
171 continue;
172 /* Some 'flags' are actually values */
173 if (flag->is_val) {
174 val = pte & flag->val;
175 if (flag->shift)
176 val = val >> flag->shift;
177 seq_printf(st->seq, " %s:%llx", flag->set, val);
178 } else {
179 if ((pte & flag->mask) == flag->val)
180 s = flag->set;
181 else
182 s = flag->clear;
183 if (s)
184 seq_printf(st->seq, " %s", s);
185 }
186 }
187}
188
189static void dump_hpte_info(struct pg_state *st, unsigned long ea, u64 v, u64 r,
190 unsigned long rpn, int bps, int aps, unsigned long lp)
191{
192 int aps_index;
193
194 while (ea >= st->marker[1].start_address) {
195 st->marker++;
196 seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
197 }
198 seq_printf(st->seq, "0x%lx:\t", ea);
199 seq_printf(st->seq, "AVPN:%llx\t", HPTE_V_AVPN_VAL(v));
200 dump_flag_info(st, v_flag_array, v, ARRAY_SIZE(v_flag_array));
201 seq_printf(st->seq, " rpn: %lx\t", rpn);
202 dump_flag_info(st, r_flag_array, r, ARRAY_SIZE(r_flag_array));
203
204 calculate_pagesize(st, bps, "base");
205 aps_index = calculate_pagesize(st, aps, "actual");
206 if (aps_index != 2)
207 seq_printf(st->seq, "LP enc: %lx", lp);
208 seq_putc(st->seq, '\n');
209}
210
211
212static int native_find(unsigned long ea, int psize, bool primary, u64 *v, u64
213 *r)
214{
215 struct hash_pte *hptep;
216 unsigned long hash, vsid, vpn, hpte_group, want_v, hpte_v;
217 int i, ssize = mmu_kernel_ssize;
218 unsigned long shift = mmu_psize_defs[psize].shift;
219
220 /* calculate hash */
221 vsid = get_kernel_vsid(ea, ssize);
222 vpn = hpt_vpn(ea, vsid, ssize);
223 hash = hpt_hash(vpn, shift, ssize);
224 want_v = hpte_encode_avpn(vpn, psize, ssize);
225
226 /* to check in the secondary hash table, we invert the hash */
227 if (!primary)
228 hash = ~hash;
229 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
230 for (i = 0; i < HPTES_PER_GROUP; i++) {
231 hptep = htab_address + hpte_group;
232 hpte_v = be64_to_cpu(hptep->v);
233
234 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
235 /* HPTE matches */
236 *v = be64_to_cpu(hptep->v);
237 *r = be64_to_cpu(hptep->r);
238 return 0;
239 }
240 ++hpte_group;
241 }
242 return -1;
243}
244
245#ifdef CONFIG_PPC_PSERIES
246static int pseries_find(unsigned long ea, int psize, bool primary, u64 *v, u64 *r)
247{
248 struct hash_pte ptes[4];
249 unsigned long vsid, vpn, hash, hpte_group, want_v;
250 int i, j, ssize = mmu_kernel_ssize;
251 long lpar_rc = 0;
252 unsigned long shift = mmu_psize_defs[psize].shift;
253
254 /* calculate hash */
255 vsid = get_kernel_vsid(ea, ssize);
256 vpn = hpt_vpn(ea, vsid, ssize);
257 hash = hpt_hash(vpn, shift, ssize);
258 want_v = hpte_encode_avpn(vpn, psize, ssize);
259
260 /* to check in the secondary hash table, we invert the hash */
261 if (!primary)
262 hash = ~hash;
263 hpte_group = ((hash & htab_hash_mask) * HPTES_PER_GROUP) & ~0x7UL;
264 /* see if we can find an entry in the hpte with this hash */
265 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
266 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
267
268 if (lpar_rc != H_SUCCESS)
269 continue;
270 for (j = 0; j < 4; j++) {
271 if (HPTE_V_COMPARE(ptes[j].v, want_v) &&
272 (ptes[j].v & HPTE_V_VALID)) {
273 /* HPTE matches */
274 *v = ptes[j].v;
275 *r = ptes[j].r;
276 return 0;
277 }
278 }
279 }
280 return -1;
281}
282#endif
283
284static void decode_r(int bps, unsigned long r, unsigned long *rpn, int *aps,
285 unsigned long *lp_bits)
286{
287 struct mmu_psize_def entry;
288 unsigned long arpn, mask, lp;
289 int penc = -2, idx = 0, shift;
290
291 /*.
292 * The LP field has 8 bits. Depending on the actual page size, some of
293 * these bits are concatenated with the APRN to get the RPN. The rest
294 * of the bits in the LP field is the LP value and is an encoding for
295 * the base page size and the actual page size.
296 *
297 * - find the mmu entry for our base page size
298 * - go through all page encodings and use the associated mask to
299 * find an encoding that matches our encoding in the LP field.
300 */
301 arpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
302 lp = arpn & 0xff;
303
304 entry = mmu_psize_defs[bps];
305 while (idx < MMU_PAGE_COUNT) {
306 penc = entry.penc[idx];
307 if ((penc != -1) && (mmu_psize_defs[idx].shift)) {
308 shift = mmu_psize_defs[idx].shift - HPTE_R_RPN_SHIFT;
309 mask = (0x1 << (shift)) - 1;
310 if ((lp & mask) == penc) {
311 *aps = mmu_psize_to_shift(idx);
312 *lp_bits = lp & mask;
313 *rpn = arpn >> shift;
314 return;
315 }
316 }
317 idx++;
318 }
319}
320
321static int base_hpte_find(unsigned long ea, int psize, bool primary, u64 *v,
322 u64 *r)
323{
324#ifdef CONFIG_PPC_PSERIES
325 if (firmware_has_feature(FW_FEATURE_LPAR))
326 return pseries_find(ea, psize, primary, v, r);
327#endif
328 return native_find(ea, psize, primary, v, r);
329}
330
331static unsigned long hpte_find(struct pg_state *st, unsigned long ea, int psize)
332{
333 unsigned long slot;
334 u64 v = 0, r = 0;
335 unsigned long rpn, lp_bits;
336 int base_psize = 0, actual_psize = 0;
337
338 if (ea < PAGE_OFFSET)
339 return -1;
340
341 /* Look in primary table */
342 slot = base_hpte_find(ea, psize, true, &v, &r);
343
344 /* Look in secondary table */
345 if (slot == -1)
346 slot = base_hpte_find(ea, psize, true, &v, &r);
347
348 /* No entry found */
349 if (slot == -1)
350 return -1;
351
352 /*
353 * We found an entry in the hash page table:
354 * - check that this has the same base page
355 * - find the actual page size
356 * - find the RPN
357 */
358 base_psize = mmu_psize_to_shift(psize);
359
360 if ((v & HPTE_V_LARGE) == HPTE_V_LARGE) {
361 decode_r(psize, r, &rpn, &actual_psize, &lp_bits);
362 } else {
363 /* 4K actual page size */
364 actual_psize = 12;
365 rpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
366 /* In this case there are no LP bits */
367 lp_bits = -1;
368 }
369 /*
370 * We didn't find a matching encoding, so the PTE we found isn't for
371 * this address.
372 */
373 if (actual_psize == -1)
374 return -1;
375
376 dump_hpte_info(st, ea, v, r, rpn, base_psize, actual_psize, lp_bits);
377 return 0;
378}
379
380static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
381{
382 pte_t *pte = pte_offset_kernel(pmd, 0);
383 unsigned long addr, pteval, psize;
384 int i, status;
385
386 for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
387 addr = start + i * PAGE_SIZE;
388 pteval = pte_val(*pte);
389
390 if (addr < VMALLOC_END)
391 psize = mmu_vmalloc_psize;
392 else
393 psize = mmu_io_psize;
394#ifdef CONFIG_PPC_64K_PAGES
395 /* check for secret 4K mappings */
396 if (((pteval & H_PAGE_COMBO) == H_PAGE_COMBO) ||
397 ((pteval & H_PAGE_4K_PFN) == H_PAGE_4K_PFN))
398 psize = mmu_io_psize;
399#endif
400 /* check for hashpte */
401 status = hpte_find(st, addr, psize);
402
403 if (((pteval & H_PAGE_HASHPTE) != H_PAGE_HASHPTE)
404 && (status != -1)) {
405 /* found a hpte that is not in the linux page tables */
406 seq_printf(st->seq, "page probably bolted before linux"
407 " pagetables were set: addr:%lx, pteval:%lx\n",
408 addr, pteval);
409 }
410 }
411}
412
413static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
414{
415 pmd_t *pmd = pmd_offset(pud, 0);
416 unsigned long addr;
417 unsigned int i;
418
419 for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
420 addr = start + i * PMD_SIZE;
421 if (!pmd_none(*pmd))
422 /* pmd exists */
423 walk_pte(st, pmd, addr);
424 }
425}
426
427static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start)
428{
429 pud_t *pud = pud_offset(pgd, 0);
430 unsigned long addr;
431 unsigned int i;
432
433 for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
434 addr = start + i * PUD_SIZE;
435 if (!pud_none(*pud))
436 /* pud exists */
437 walk_pmd(st, pud, addr);
438 }
439}
440
441static void walk_pagetables(struct pg_state *st)
442{
443 pgd_t *pgd = pgd_offset_k(0UL);
444 unsigned int i;
445 unsigned long addr;
446
447 /*
448 * Traverse the linux pagetable structure and dump pages that are in
449 * the hash pagetable.
450 */
451 for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
452 addr = KERN_VIRT_START + i * PGDIR_SIZE;
453 if (!pgd_none(*pgd))
454 /* pgd exists */
455 walk_pud(st, pgd, addr);
456 }
457}
458
459
460static void walk_linearmapping(struct pg_state *st)
461{
462 unsigned long addr;
463
464 /*
465 * Traverse the linear mapping section of virtual memory and dump pages
466 * that are in the hash pagetable.
467 */
468 unsigned long psize = 1 << mmu_psize_defs[mmu_linear_psize].shift;
469
470 for (addr = PAGE_OFFSET; addr < PAGE_OFFSET +
471 memblock_end_of_DRAM(); addr += psize)
472 hpte_find(st, addr, mmu_linear_psize);
473}
474
475static void walk_vmemmap(struct pg_state *st)
476{
477#ifdef CONFIG_SPARSEMEM_VMEMMAP
478 struct vmemmap_backing *ptr = vmemmap_list;
479
480 /*
481 * Traverse the vmemmaped memory and dump pages that are in the hash
482 * pagetable.
483 */
484 while (ptr->list) {
485 hpte_find(st, ptr->virt_addr, mmu_vmemmap_psize);
486 ptr = ptr->list;
487 }
488 seq_puts(st->seq, "---[ vmemmap end ]---\n");
489#endif
490}
491
492static void populate_markers(void)
493{
494 address_markers[0].start_address = PAGE_OFFSET;
495 address_markers[1].start_address = VMALLOC_START;
496 address_markers[2].start_address = VMALLOC_END;
497 address_markers[3].start_address = ISA_IO_BASE;
498 address_markers[4].start_address = ISA_IO_END;
499 address_markers[5].start_address = PHB_IO_BASE;
500 address_markers[6].start_address = PHB_IO_END;
501 address_markers[7].start_address = IOREMAP_BASE;
502 address_markers[8].start_address = IOREMAP_END;
503#ifdef CONFIG_PPC_BOOK3S_64
504 address_markers[9].start_address = H_VMEMMAP_BASE;
505#else
506 address_markers[9].start_address = VMEMMAP_BASE;
507#endif
508}
509
510static int ptdump_show(struct seq_file *m, void *v)
511{
512 struct pg_state st = {
513 .seq = m,
514 .start_address = PAGE_OFFSET,
515 .marker = address_markers,
516 };
517 /*
518 * Traverse the 0xc, 0xd and 0xf areas of the kernel virtual memory and
519 * dump pages that are in the hash pagetable.
520 */
521 walk_linearmapping(&st);
522 walk_pagetables(&st);
523 walk_vmemmap(&st);
524 return 0;
525}
526
527static int ptdump_open(struct inode *inode, struct file *file)
528{
529 return single_open(file, ptdump_show, NULL);
530}
531
532static const struct file_operations ptdump_fops = {
533 .open = ptdump_open,
534 .read = seq_read,
535 .llseek = seq_lseek,
536 .release = single_release,
537};
538
539static int ptdump_init(void)
540{
541 struct dentry *debugfs_file;
542
543 if (!radix_enabled()) {
544 populate_markers();
545 debugfs_file = debugfs_create_file("kernel_hash_pagetable",
546 0400, NULL, NULL, &ptdump_fops);
547 return debugfs_file ? 0 : -ENOMEM;
548 }
549 return 0;
550}
551device_initcall(ptdump_init);