2 * Copyright 2016, Rashmica Gupta, IBM Corp.
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.
8 * If radix is enabled then there is no hash page table and so no debugfs file
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
16 #include <linux/debugfs.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>
26 #include <asm/pgalloc.h>
27 #include <asm/plpar_wrappers.h>
28 #include <linux/memblock.h>
29 #include <asm/firmware.h>
33 const struct addr_marker *marker;
34 unsigned long start_address;
40 unsigned long start_address;
44 static 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" },
50 { 0, "phb I/O start" },
52 { 0, "I/O remap start" },
53 { 0, "I/O remap end" },
54 { 0, "vmemmap start" },
67 static const struct flag_info v_flag_array[] = {
70 .val = SLB_VSID_B_256M,
72 .clear = "ssize: 1T ",
74 .mask = HPTE_V_SECONDARY,
75 .val = HPTE_V_SECONDARY,
84 .mask = HPTE_V_BOLTED,
91 static const struct flag_info r_flag_array[] = {
93 .mask = HPTE_R_PP0 | HPTE_R_PP,
97 .mask = HPTE_R_PP0 | HPTE_R_PP,
101 .mask = HPTE_R_PP0 | HPTE_R_PP,
105 .mask = HPTE_R_PP0 | HPTE_R_PP,
109 .mask = HPTE_R_PP0 | HPTE_R_PP,
113 .mask = HPTE_R_KEY_HI | HPTE_R_KEY_LO,
114 .val = HPTE_R_KEY_HI | HPTE_R_KEY_LO,
147 static int calculate_pagesize(struct pg_state *st, int ps, char s[])
149 static const char units[] = "BKMGTPE";
150 const char *unit = units;
152 while (ps > 9 && unit[1]) {
156 seq_printf(st->seq, " %s_ps: %i%c\t", s, 1<<ps, *unit);
160 static void dump_flag_info(struct pg_state *st, const struct flag_info
161 *flag, u64 pte, int num)
165 for (i = 0; i < num; i++, flag++) {
166 const char *s = NULL;
169 /* flag not defined so don't check it */
172 /* Some 'flags' are actually values */
174 val = pte & flag->val;
176 val = val >> flag->shift;
177 seq_printf(st->seq, " %s:%llx", flag->set, val);
179 if ((pte & flag->mask) == flag->val)
184 seq_printf(st->seq, " %s", s);
189 static 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)
194 while (ea >= st->marker[1].start_address) {
196 seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
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));
204 calculate_pagesize(st, bps, "base");
205 aps_index = calculate_pagesize(st, aps, "actual");
207 seq_printf(st->seq, "LP enc: %lx", lp);
208 seq_puts(st->seq, "\n");
212 static int native_find(unsigned long ea, int psize, bool primary, u64 *v, u64
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;
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);
226 /* to check in the secondary hash table, we invert the 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);
234 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
236 *v = be64_to_cpu(hptep->v);
237 *r = be64_to_cpu(hptep->r);
245 #ifdef CONFIG_PPC_PSERIES
246 static int pseries_find(unsigned long ea, int psize, bool primary, u64 *v, u64 *r)
248 struct hash_pte ptes[4];
249 unsigned long vsid, vpn, hash, hpte_group, want_v;
250 int i, j, ssize = mmu_kernel_ssize;
252 unsigned long shift = mmu_psize_defs[psize].shift;
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);
260 /* to check in the secondary hash table, we invert the 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);
268 if (lpar_rc != H_SUCCESS)
270 for (j = 0; j < 4; j++) {
271 if (HPTE_V_COMPARE(ptes[j].v, want_v) &&
272 (ptes[j].v & HPTE_V_VALID)) {
284 static void decode_r(int bps, unsigned long r, unsigned long *rpn, int *aps,
285 unsigned long *lp_bits)
287 struct mmu_psize_def entry;
288 unsigned long arpn, mask, lp;
289 int penc = -2, idx = 0, shift;
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.
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.
301 arpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
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;
321 static int base_hpte_find(unsigned long ea, int psize, bool primary, u64 *v,
324 #ifdef CONFIG_PPC_PSERIES
325 if (firmware_has_feature(FW_FEATURE_LPAR))
326 return pseries_find(ea, psize, primary, v, r);
328 return native_find(ea, psize, primary, v, r);
331 static unsigned long hpte_find(struct pg_state *st, unsigned long ea, int psize)
335 unsigned long rpn, lp_bits;
336 int base_psize = 0, actual_psize = 0;
338 if (ea <= PAGE_OFFSET)
341 /* Look in primary table */
342 slot = base_hpte_find(ea, psize, true, &v, &r);
344 /* Look in secondary table */
346 slot = base_hpte_find(ea, psize, true, &v, &r);
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
358 base_psize = mmu_psize_to_shift(psize);
360 if ((v & HPTE_V_LARGE) == HPTE_V_LARGE) {
361 decode_r(psize, r, &rpn, &actual_psize, &lp_bits);
363 /* 4K actual page size */
365 rpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
366 /* In this case there are no LP bits */
370 * We didn't find a matching encoding, so the PTE we found isn't for
373 if (actual_psize == -1)
376 dump_hpte_info(st, ea, v, r, rpn, base_psize, actual_psize, lp_bits);
380 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
382 pte_t *pte = pte_offset_kernel(pmd, 0);
383 unsigned long addr, pteval, psize;
386 for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
387 addr = start + i * PAGE_SIZE;
388 pteval = pte_val(*pte);
390 if (addr < VMALLOC_END)
391 psize = mmu_vmalloc_psize;
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;
400 /* check for hashpte */
401 status = hpte_find(st, addr, psize);
403 if (((pteval & H_PAGE_HASHPTE) != H_PAGE_HASHPTE)
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",
413 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
415 pmd_t *pmd = pmd_offset(pud, 0);
419 for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
420 addr = start + i * PMD_SIZE;
423 walk_pte(st, pmd, addr);
427 static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start)
429 pud_t *pud = pud_offset(pgd, 0);
433 for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
434 addr = start + i * PUD_SIZE;
437 walk_pmd(st, pud, addr);
441 static void walk_pagetables(struct pg_state *st)
443 pgd_t *pgd = pgd_offset_k(0UL);
448 * Traverse the linux pagetable structure and dump pages that are in
449 * the hash pagetable.
451 for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
452 addr = KERN_VIRT_START + i * PGDIR_SIZE;
455 walk_pud(st, pgd, addr);
460 static void walk_linearmapping(struct pg_state *st)
465 * Traverse the linear mapping section of virtual memory and dump pages
466 * that are in the hash pagetable.
468 unsigned long psize = 1 << mmu_psize_defs[mmu_linear_psize].shift;
470 for (addr = PAGE_OFFSET; addr < PAGE_OFFSET +
471 memblock_phys_mem_size(); addr += psize)
472 hpte_find(st, addr, mmu_linear_psize);
475 static void walk_vmemmap(struct pg_state *st)
477 #ifdef CONFIG_SPARSEMEM_VMEMMAP
478 struct vmemmap_backing *ptr = vmemmap_list;
481 * Traverse the vmemmaped memory and dump pages that are in the hash
485 hpte_find(st, ptr->virt_addr, mmu_vmemmap_psize);
488 seq_puts(st->seq, "---[ vmemmap end ]---\n");
492 static void populate_markers(void)
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_STD_MMU_64
504 address_markers[9].start_address = H_VMEMMAP_BASE;
506 address_markers[9].start_address = VMEMMAP_BASE;
510 static int ptdump_show(struct seq_file *m, void *v)
512 struct pg_state st = {
514 .start_address = PAGE_OFFSET,
515 .marker = address_markers,
518 * Traverse the 0xc, 0xd and 0xf areas of the kernel virtual memory and
519 * dump pages that are in the hash pagetable.
521 walk_linearmapping(&st);
522 walk_pagetables(&st);
527 static int ptdump_open(struct inode *inode, struct file *file)
529 return single_open(file, ptdump_show, NULL);
532 static const struct file_operations ptdump_fops = {
536 .release = single_release,
539 static int ptdump_init(void)
541 struct dentry *debugfs_file;
543 if (!radix_enabled()) {
545 debugfs_file = debugfs_create_file("kernel_hash_pagetable",
546 0400, NULL, NULL, &ptdump_fops);
547 return debugfs_file ? 0 : -ENOMEM;
551 device_initcall(ptdump_init);