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
calculate_pagesize(struct pg_state * st,int ps,char s[])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
dump_flag_info(struct pg_state * st,const struct flag_info * flag,u64 pte,int num)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
dump_hpte_info(struct pg_state * st,unsigned long ea,u64 v,u64 r,unsigned long rpn,int bps,int aps,unsigned long lp)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
native_find(unsigned long ea,int psize,bool primary,u64 * v,u64 * r)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
pseries_find(unsigned long ea,int psize,bool primary,u64 * v,u64 * r)239 static int pseries_find(unsigned long ea, int psize, bool primary, u64 *v, u64 *r)
240 {
241 struct hash_pte ptes[4];
242 unsigned long vsid, vpn, hash, hpte_group, want_v;
243 int i, j, ssize = mmu_kernel_ssize;
244 long lpar_rc = 0;
245 unsigned long shift = mmu_psize_defs[psize].shift;
246
247 /* calculate hash */
248 vsid = get_kernel_vsid(ea, ssize);
249 vpn = hpt_vpn(ea, vsid, ssize);
250 hash = hpt_hash(vpn, shift, ssize);
251 want_v = hpte_encode_avpn(vpn, psize, ssize);
252
253 /* to check in the secondary hash table, we invert the hash */
254 if (!primary)
255 hash = ~hash;
256 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
257 /* see if we can find an entry in the hpte with this hash */
258 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
259 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
260
261 if (lpar_rc)
262 continue;
263 for (j = 0; j < 4; j++) {
264 if (HPTE_V_COMPARE(ptes[j].v, want_v) &&
265 (ptes[j].v & HPTE_V_VALID)) {
266 /* HPTE matches */
267 *v = ptes[j].v;
268 *r = ptes[j].r;
269 return 0;
270 }
271 }
272 }
273 return -1;
274 }
275
decode_r(int bps,unsigned long r,unsigned long * rpn,int * aps,unsigned long * lp_bits)276 static void decode_r(int bps, unsigned long r, unsigned long *rpn, int *aps,
277 unsigned long *lp_bits)
278 {
279 struct mmu_psize_def entry;
280 unsigned long arpn, mask, lp;
281 int penc = -2, idx = 0, shift;
282
283 /*.
284 * The LP field has 8 bits. Depending on the actual page size, some of
285 * these bits are concatenated with the APRN to get the RPN. The rest
286 * of the bits in the LP field is the LP value and is an encoding for
287 * the base page size and the actual page size.
288 *
289 * - find the mmu entry for our base page size
290 * - go through all page encodings and use the associated mask to
291 * find an encoding that matches our encoding in the LP field.
292 */
293 arpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
294 lp = arpn & 0xff;
295
296 entry = mmu_psize_defs[bps];
297 while (idx < MMU_PAGE_COUNT) {
298 penc = entry.penc[idx];
299 if ((penc != -1) && (mmu_psize_defs[idx].shift)) {
300 shift = mmu_psize_defs[idx].shift - HPTE_R_RPN_SHIFT;
301 mask = (0x1 << (shift)) - 1;
302 if ((lp & mask) == penc) {
303 *aps = mmu_psize_to_shift(idx);
304 *lp_bits = lp & mask;
305 *rpn = arpn >> shift;
306 return;
307 }
308 }
309 idx++;
310 }
311 }
312
base_hpte_find(unsigned long ea,int psize,bool primary,u64 * v,u64 * r)313 static int base_hpte_find(unsigned long ea, int psize, bool primary, u64 *v,
314 u64 *r)
315 {
316 if (IS_ENABLED(CONFIG_PPC_PSERIES) && firmware_has_feature(FW_FEATURE_LPAR))
317 return pseries_find(ea, psize, primary, v, r);
318
319 return native_find(ea, psize, primary, v, r);
320 }
321
hpte_find(struct pg_state * st,unsigned long ea,int psize)322 static unsigned long hpte_find(struct pg_state *st, unsigned long ea, int psize)
323 {
324 unsigned long slot;
325 u64 v = 0, r = 0;
326 unsigned long rpn, lp_bits;
327 int base_psize = 0, actual_psize = 0;
328
329 if (ea < PAGE_OFFSET)
330 return -1;
331
332 /* Look in primary table */
333 slot = base_hpte_find(ea, psize, true, &v, &r);
334
335 /* Look in secondary table */
336 if (slot == -1)
337 slot = base_hpte_find(ea, psize, false, &v, &r);
338
339 /* No entry found */
340 if (slot == -1)
341 return -1;
342
343 /*
344 * We found an entry in the hash page table:
345 * - check that this has the same base page
346 * - find the actual page size
347 * - find the RPN
348 */
349 base_psize = mmu_psize_to_shift(psize);
350
351 if ((v & HPTE_V_LARGE) == HPTE_V_LARGE) {
352 decode_r(psize, r, &rpn, &actual_psize, &lp_bits);
353 } else {
354 /* 4K actual page size */
355 actual_psize = 12;
356 rpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
357 /* In this case there are no LP bits */
358 lp_bits = -1;
359 }
360 /*
361 * We didn't find a matching encoding, so the PTE we found isn't for
362 * this address.
363 */
364 if (actual_psize == -1)
365 return -1;
366
367 dump_hpte_info(st, ea, v, r, rpn, base_psize, actual_psize, lp_bits);
368 return 0;
369 }
370
walk_pte(struct pg_state * st,pmd_t * pmd,unsigned long start)371 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
372 {
373 pte_t *pte = pte_offset_kernel(pmd, 0);
374 unsigned long addr, pteval, psize;
375 int i, status;
376
377 for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
378 addr = start + i * PAGE_SIZE;
379 pteval = pte_val(*pte);
380
381 if (addr < VMALLOC_END)
382 psize = mmu_vmalloc_psize;
383 else
384 psize = mmu_io_psize;
385
386 /* check for secret 4K mappings */
387 if (IS_ENABLED(CONFIG_PPC_64K_PAGES) &&
388 ((pteval & H_PAGE_COMBO) == H_PAGE_COMBO ||
389 (pteval & H_PAGE_4K_PFN) == H_PAGE_4K_PFN))
390 psize = mmu_io_psize;
391
392 /* check for hashpte */
393 status = hpte_find(st, addr, psize);
394
395 if (((pteval & H_PAGE_HASHPTE) != H_PAGE_HASHPTE)
396 && (status != -1)) {
397 /* found a hpte that is not in the linux page tables */
398 seq_printf(st->seq, "page probably bolted before linux"
399 " pagetables were set: addr:%lx, pteval:%lx\n",
400 addr, pteval);
401 }
402 }
403 }
404
walk_pmd(struct pg_state * st,pud_t * pud,unsigned long start)405 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
406 {
407 pmd_t *pmd = pmd_offset(pud, 0);
408 unsigned long addr;
409 unsigned int i;
410
411 for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
412 addr = start + i * PMD_SIZE;
413 if (!pmd_none(*pmd))
414 /* pmd exists */
415 walk_pte(st, pmd, addr);
416 }
417 }
418
walk_pud(struct pg_state * st,p4d_t * p4d,unsigned long start)419 static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
420 {
421 pud_t *pud = pud_offset(p4d, 0);
422 unsigned long addr;
423 unsigned int i;
424
425 for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
426 addr = start + i * PUD_SIZE;
427 if (!pud_none(*pud))
428 /* pud exists */
429 walk_pmd(st, pud, addr);
430 }
431 }
432
walk_p4d(struct pg_state * st,pgd_t * pgd,unsigned long start)433 static void walk_p4d(struct pg_state *st, pgd_t *pgd, unsigned long start)
434 {
435 p4d_t *p4d = p4d_offset(pgd, 0);
436 unsigned long addr;
437 unsigned int i;
438
439 for (i = 0; i < PTRS_PER_P4D; i++, p4d++) {
440 addr = start + i * P4D_SIZE;
441 if (!p4d_none(*p4d))
442 /* p4d exists */
443 walk_pud(st, p4d, addr);
444 }
445 }
446
walk_pagetables(struct pg_state * st)447 static void walk_pagetables(struct pg_state *st)
448 {
449 pgd_t *pgd = pgd_offset_k(0UL);
450 unsigned int i;
451 unsigned long addr;
452
453 /*
454 * Traverse the linux pagetable structure and dump pages that are in
455 * the hash pagetable.
456 */
457 for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
458 addr = KERN_VIRT_START + i * PGDIR_SIZE;
459 if (!pgd_none(*pgd))
460 /* pgd exists */
461 walk_p4d(st, pgd, addr);
462 }
463 }
464
465
walk_linearmapping(struct pg_state * st)466 static void walk_linearmapping(struct pg_state *st)
467 {
468 unsigned long addr;
469
470 /*
471 * Traverse the linear mapping section of virtual memory and dump pages
472 * that are in the hash pagetable.
473 */
474 unsigned long psize = 1 << mmu_psize_defs[mmu_linear_psize].shift;
475
476 for (addr = PAGE_OFFSET; addr < PAGE_OFFSET +
477 memblock_end_of_DRAM(); addr += psize)
478 hpte_find(st, addr, mmu_linear_psize);
479 }
480
walk_vmemmap(struct pg_state * st)481 static void walk_vmemmap(struct pg_state *st)
482 {
483 struct vmemmap_backing *ptr = vmemmap_list;
484
485 if (!IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
486 return;
487 /*
488 * Traverse the vmemmaped memory and dump pages that are in the hash
489 * pagetable.
490 */
491 while (ptr->list) {
492 hpte_find(st, ptr->virt_addr, mmu_vmemmap_psize);
493 ptr = ptr->list;
494 }
495 seq_puts(st->seq, "---[ vmemmap end ]---\n");
496 }
497
populate_markers(void)498 static void populate_markers(void)
499 {
500 address_markers[0].start_address = PAGE_OFFSET;
501 address_markers[1].start_address = VMALLOC_START;
502 address_markers[2].start_address = VMALLOC_END;
503 address_markers[3].start_address = ISA_IO_BASE;
504 address_markers[4].start_address = ISA_IO_END;
505 address_markers[5].start_address = PHB_IO_BASE;
506 address_markers[6].start_address = PHB_IO_END;
507 address_markers[7].start_address = IOREMAP_BASE;
508 address_markers[8].start_address = IOREMAP_END;
509 address_markers[9].start_address = H_VMEMMAP_START;
510 }
511
ptdump_show(struct seq_file * m,void * v)512 static int ptdump_show(struct seq_file *m, void *v)
513 {
514 struct pg_state st = {
515 .seq = m,
516 .start_address = PAGE_OFFSET,
517 .marker = address_markers,
518 };
519 /*
520 * Traverse the 0xc, 0xd and 0xf areas of the kernel virtual memory and
521 * dump pages that are in the hash pagetable.
522 */
523 walk_linearmapping(&st);
524 walk_pagetables(&st);
525 walk_vmemmap(&st);
526 return 0;
527 }
528
ptdump_open(struct inode * inode,struct file * file)529 static int ptdump_open(struct inode *inode, struct file *file)
530 {
531 return single_open(file, ptdump_show, NULL);
532 }
533
534 static const struct file_operations ptdump_fops = {
535 .open = ptdump_open,
536 .read = seq_read,
537 .llseek = seq_lseek,
538 .release = single_release,
539 };
540
ptdump_init(void)541 static int ptdump_init(void)
542 {
543 if (!radix_enabled()) {
544 populate_markers();
545 debugfs_create_file("kernel_hash_pagetable", 0400, NULL, NULL,
546 &ptdump_fops);
547 }
548 return 0;
549 }
550 device_initcall(ptdump_init);
551