hashpagetable.c

  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
 25struct 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
 33struct addr_marker {
 34	unsigned long start_address;
 35	const char *name;
 36};
 37
 38static 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
 52struct 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
 61static 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
 85static 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
141static 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
154static 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
183static 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
206static 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	if (cpu_has_feature(CPU_FTR_ARCH_300))
220		want_v = hpte_old_to_new_v(want_v);
221
222	/* to check in the secondary hash table, we invert the hash */
223	if (!primary)
224		hash = ~hash;
225	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
226	for (i = 0; i < HPTES_PER_GROUP; i++) {
227		hptep = htab_address + hpte_group;
228		hpte_v = be64_to_cpu(hptep->v);
229
230		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
231			/* HPTE matches */
232			*v = be64_to_cpu(hptep->v);
233			*r = be64_to_cpu(hptep->r);
234			if (cpu_has_feature(CPU_FTR_ARCH_300)) {
235				*v = hpte_new_to_old_v(*v, *r);
236				*r = hpte_new_to_old_r(*r);
237			}
238			return 0;
239		}
240		++hpte_group;
241	}
242	return -1;
243}
244
245static int pseries_find(unsigned long ea, int psize, bool primary, u64 *v, u64 *r)
246{
247	struct {
248		unsigned long v;
249		unsigned long r;
250	} ptes[4];
251	unsigned long vsid, vpn, hash, hpte_group, want_v;
252	int i, j, ssize = mmu_kernel_ssize;
253	long lpar_rc = 0;
254	unsigned long shift = mmu_psize_defs[psize].shift;
255
256	/* calculate hash */
257	vsid = get_kernel_vsid(ea, ssize);
258	vpn  = hpt_vpn(ea, vsid, ssize);
259	hash = hpt_hash(vpn, shift, ssize);
260	want_v = hpte_encode_avpn(vpn, psize, ssize);
261
262	/* to check in the secondary hash table, we invert the hash */
263	if (!primary)
264		hash = ~hash;
265	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
266	/* see if we can find an entry in the hpte with this hash */
267	for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
268		lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
269
270		if (lpar_rc)
271			continue;
272		for (j = 0; j < 4; j++) {
273			if (HPTE_V_COMPARE(ptes[j].v, want_v) &&
274					(ptes[j].v & HPTE_V_VALID)) {
275				/* HPTE matches */
276				*v = ptes[j].v;
277				*r = ptes[j].r;
278				return 0;
279			}
280		}
281	}
282	return -1;
283}
284
285static void decode_r(int bps, unsigned long r, unsigned long *rpn, int *aps,
286		unsigned long *lp_bits)
287{
288	struct mmu_psize_def entry;
289	unsigned long arpn, mask, lp;
290	int penc = -2, idx = 0, shift;
291
292	/*.
293	 * The LP field has 8 bits. Depending on the actual page size, some of
294	 * these bits are concatenated with the APRN to get the RPN. The rest
295	 * of the bits in the LP field is the LP value and is an encoding for
296	 * the base page size and the actual page size.
297	 *
298	 *  -	find the mmu entry for our base page size
299	 *  -	go through all page encodings and use the associated mask to
300	 *	find an encoding that matches our encoding in the LP field.
301	 */
302	arpn = (r & HPTE_R_RPN) >> HPTE_R_RPN_SHIFT;
303	lp = arpn & 0xff;
304
305	entry = mmu_psize_defs[bps];
306	while (idx < MMU_PAGE_COUNT) {
307		penc = entry.penc[idx];
308		if ((penc != -1) && (mmu_psize_defs[idx].shift)) {
309			shift = mmu_psize_defs[idx].shift -  HPTE_R_RPN_SHIFT;
310			mask = (0x1 << (shift)) - 1;
311			if ((lp & mask) == penc) {
312				*aps = mmu_psize_to_shift(idx);
313				*lp_bits = lp & mask;
314				*rpn = arpn >> shift;
315				return;
316			}
317		}
318		idx++;
319	}
320}
321
322static int base_hpte_find(unsigned long ea, int psize, bool primary, u64 *v,
323			  u64 *r)
324{
325	if (IS_ENABLED(CONFIG_PPC_PSERIES) && firmware_has_feature(FW_FEATURE_LPAR))
326		return pseries_find(ea, psize, primary, v, r);
327
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, false, &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
395		/* check for secret 4K mappings */
396		if (IS_ENABLED(CONFIG_PPC_64K_PAGES) &&
397		    ((pteval & H_PAGE_COMBO) == H_PAGE_COMBO ||
398		     (pteval & H_PAGE_4K_PFN) == H_PAGE_4K_PFN))
399			psize = mmu_io_psize;
400
401		/* check for hashpte */
402		status = hpte_find(st, addr, psize);
403
404		if (((pteval & H_PAGE_HASHPTE) != H_PAGE_HASHPTE)
405				&& (status != -1)) {
406		/* found a hpte that is not in the linux page tables */
407			seq_printf(st->seq, "page probably bolted before linux"
408				" pagetables were set: addr:%lx, pteval:%lx\n",
409				addr, pteval);
410		}
411	}
412}
413
414static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
415{
416	pmd_t *pmd = pmd_offset(pud, 0);
417	unsigned long addr;
418	unsigned int i;
419
420	for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
421		addr = start + i * PMD_SIZE;
422		if (!pmd_none(*pmd))
423			/* pmd exists */
424			walk_pte(st, pmd, addr);
425	}
426}
427
428static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
429{
430	pud_t *pud = pud_offset(p4d, 0);
431	unsigned long addr;
432	unsigned int i;
433
434	for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
435		addr = start + i * PUD_SIZE;
436		if (!pud_none(*pud))
437			/* pud exists */
438			walk_pmd(st, pud, addr);
439	}
440}
441
442static void walk_p4d(struct pg_state *st, pgd_t *pgd, unsigned long start)
443{
444	p4d_t *p4d = p4d_offset(pgd, 0);
445	unsigned long addr;
446	unsigned int i;
447
448	for (i = 0; i < PTRS_PER_P4D; i++, p4d++) {
449		addr = start + i * P4D_SIZE;
450		if (!p4d_none(*p4d))
451			/* p4d exists */
452			walk_pud(st, p4d, addr);
453	}
454}
455
456static void walk_pagetables(struct pg_state *st)
457{
458	pgd_t *pgd = pgd_offset_k(0UL);
459	unsigned int i;
460	unsigned long addr;
461
462	/*
463	 * Traverse the linux pagetable structure and dump pages that are in
464	 * the hash pagetable.
465	 */
466	for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
467		addr = KERN_VIRT_START + i * PGDIR_SIZE;
468		if (!pgd_none(*pgd))
469			/* pgd exists */
470			walk_p4d(st, pgd, addr);
471	}
472}
473
474
475static void walk_linearmapping(struct pg_state *st)
476{
477	unsigned long addr;
478
479	/*
480	 * Traverse the linear mapping section of virtual memory and dump pages
481	 * that are in the hash pagetable.
482	 */
483	unsigned long psize = 1 << mmu_psize_defs[mmu_linear_psize].shift;
484
485	for (addr = PAGE_OFFSET; addr < PAGE_OFFSET +
486			memblock_end_of_DRAM(); addr += psize)
487		hpte_find(st, addr, mmu_linear_psize);
488}
489
490static void walk_vmemmap(struct pg_state *st)
491{
492	struct vmemmap_backing *ptr = vmemmap_list;
493
494	if (!IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
495		return;
496	/*
497	 * Traverse the vmemmaped memory and dump pages that are in the hash
498	 * pagetable.
499	 */
500	while (ptr) {
501		hpte_find(st, ptr->virt_addr, mmu_vmemmap_psize);
502		ptr = ptr->list;
503	}
504	seq_puts(st->seq, "---[ vmemmap end ]---\n");
505}
506
507static void populate_markers(void)
508{
509	address_markers[0].start_address = PAGE_OFFSET;
510	address_markers[1].start_address = VMALLOC_START;
511	address_markers[2].start_address = VMALLOC_END;
512	address_markers[3].start_address = ISA_IO_BASE;
513	address_markers[4].start_address = ISA_IO_END;
514	address_markers[5].start_address = PHB_IO_BASE;
515	address_markers[6].start_address = PHB_IO_END;
516	address_markers[7].start_address = IOREMAP_BASE;
517	address_markers[8].start_address = IOREMAP_END;
518	address_markers[9].start_address =  H_VMEMMAP_START;
519}
520
521static int ptdump_show(struct seq_file *m, void *v)
522{
523	struct pg_state st = {
524		.seq = m,
525		.start_address = PAGE_OFFSET,
526		.marker = address_markers,
527	};
528	/*
529	 * Traverse the 0xc, 0xd and 0xf areas of the kernel virtual memory and
530	 * dump pages that are in the hash pagetable.
531	 */
532	walk_linearmapping(&st);
533	walk_pagetables(&st);
534	walk_vmemmap(&st);
535	return 0;
536}
537
538DEFINE_SHOW_ATTRIBUTE(ptdump);
539
540static int ptdump_init(void)
541{
542	if (!radix_enabled()) {
543		populate_markers();
544		debugfs_create_file("kernel_hash_pagetable", 0400, NULL, NULL,
545				    &ptdump_fops);
546	}
547	return 0;
548}
549device_initcall(ptdump_init);