powerpc/mm: Dump linux pagetables
authorRashmica Gupta <rashmicy@gmail.com>
Fri, 27 May 2016 05:48:59 +0000 (15:48 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Thu, 17 Nov 2016 06:11:46 +0000 (17:11 +1100)
Useful to be able to dump the kernels page tables to check permissions
and memory types - derived from arm64's implementation.

Add a debugfs file to check the page tables. To use this the PPC_PTDUMP
config option must be selected.

Signed-off-by: Rashmica Gupta <rashmicy@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/Kconfig.debug
arch/powerpc/mm/Makefile
arch/powerpc/mm/dump_linuxpagetables.c [new file with mode: 0644]

index 63292f64b25a384f716ae7aaf08f90d927765ca5..20cf770611ec4a9b136a5ccf4eced60b04f48224 100644 (file)
@@ -354,4 +354,16 @@ config FAIL_IOMMU
 
          If you are unsure, say N.
 
+config PPC_PTDUMP
+        bool "Export kernel pagetable layout to userspace via debugfs"
+        depends on DEBUG_KERNEL
+        select DEBUG_FS
+        help
+         This option exports the state of the kernel pagetables to a
+         debugfs file. This is only useful for kernel developers who are
+         working in architecture specific areas of the kernel - probably
+         not a good idea to enable this feature in a production kernel.
+
+         If you are unsure, say N.
+
 endmenu
index 1a4e570f789470dc1fa8108a7470bded5f648558..5b3e4d560315e165d356c53b23ef16421f3fa2c7 100644 (file)
@@ -42,3 +42,4 @@ obj-$(CONFIG_NOT_COHERENT_CACHE) += dma-noncoherent.o
 obj-$(CONFIG_HIGHMEM)          += highmem.o
 obj-$(CONFIG_PPC_COPRO_BASE)   += copro_fault.o
 obj-$(CONFIG_SPAPR_TCE_IOMMU)  += mmu_context_iommu.o
+obj-$(CONFIG_PPC_PTDUMP)       += dump_linuxpagetables.o
diff --git a/arch/powerpc/mm/dump_linuxpagetables.c b/arch/powerpc/mm/dump_linuxpagetables.c
new file mode 100644 (file)
index 0000000..d242bc7
--- /dev/null
@@ -0,0 +1,440 @@
+/*
+ * Copyright 2016, Rashmica Gupta, IBM Corp.
+ *
+ * This traverses the kernel pagetables and dumps the
+ * information about the used sections of memory to
+ * /sys/kernel/debug/kernel_pagetables.
+ *
+ * Derived from the arm64 implementation:
+ * Copyright (c) 2014, The Linux Foundation, Laura Abbott.
+ * (C) Copyright 2008 Intel Corporation, Arjan van de Ven.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+#include <linux/debugfs.h>
+#include <linux/fs.h>
+#include <linux/io.h>
+#include <linux/mm.h>
+#include <linux/sched.h>
+#include <linux/seq_file.h>
+#include <asm/fixmap.h>
+#include <asm/pgtable.h>
+#include <linux/const.h>
+#include <asm/page.h>
+#include <asm/pgalloc.h>
+
+/*
+ * To visualise what is happening,
+ *
+ *  - PTRS_PER_P** = how many entries there are in the corresponding P**
+ *  - P**_SHIFT = how many bits of the address we use to index into the
+ * corresponding P**
+ *  - P**_SIZE is how much memory we can access through the table - not the
+ * size of the table itself.
+ * P**={PGD, PUD, PMD, PTE}
+ *
+ *
+ * Each entry of the PGD points to a PUD. Each entry of a PUD points to a
+ * PMD. Each entry of a PMD points to a PTE. And every PTE entry points to
+ * a page.
+ *
+ * In the case where there are only 3 levels, the PUD is folded into the
+ * PGD: every PUD has only one entry which points to the PMD.
+ *
+ * The page dumper groups page table entries of the same type into a single
+ * description. It uses pg_state to track the range information while
+ * iterating over the PTE entries. When the continuity is broken it then
+ * dumps out a description of the range - ie PTEs that are virtually contiguous
+ * with the same PTE flags are chunked together. This is to make it clear how
+ * different areas of the kernel virtual memory are used.
+ *
+ */
+struct pg_state {
+       struct seq_file *seq;
+       const struct addr_marker *marker;
+       unsigned long start_address;
+       unsigned int level;
+       u64 current_flags;
+};
+
+struct addr_marker {
+       unsigned long start_address;
+       const char *name;
+};
+
+static struct addr_marker address_markers[] = {
+       { 0,    "Start of kernel VM" },
+       { 0,    "vmalloc() Area" },
+       { 0,    "vmalloc() End" },
+       { 0,    "isa I/O start" },
+       { 0,    "isa I/O end" },
+       { 0,    "phb I/O start" },
+       { 0,    "phb I/O end" },
+       { 0,    "I/O remap start" },
+       { 0,    "I/O remap end" },
+       { 0,    "vmemmap start" },
+       { -1,   NULL },
+};
+
+struct flag_info {
+       u64             mask;
+       u64             val;
+       const char      *set;
+       const char      *clear;
+       bool            is_val;
+       int             shift;
+};
+
+static const struct flag_info flag_array[] = {
+       {
+#ifdef CONFIG_PPC_STD_MMU_64
+               .mask   = _PAGE_PRIVILEGED,
+               .val    = 0,
+#else
+               .mask   = _PAGE_USER,
+               .val    = _PAGE_USER,
+#endif
+               .set    = "user",
+               .clear  = "    ",
+       }, {
+               .mask   = _PAGE_RW,
+               .val    = _PAGE_RW,
+               .set    = "rw",
+               .clear  = "ro",
+       }, {
+               .mask   = _PAGE_EXEC,
+               .val    = _PAGE_EXEC,
+               .set    = " X ",
+               .clear  = "   ",
+       }, {
+               .mask   = _PAGE_PTE,
+               .val    = _PAGE_PTE,
+               .set    = "pte",
+               .clear  = "   ",
+       }, {
+               .mask   = _PAGE_PRESENT,
+               .val    = _PAGE_PRESENT,
+               .set    = "present",
+               .clear  = "       ",
+       }, {
+#ifdef CONFIG_PPC_STD_MMU_64
+               .mask   = H_PAGE_HASHPTE,
+               .val    = H_PAGE_HASHPTE,
+#else
+               .mask   = _PAGE_HASHPTE,
+               .val    = _PAGE_HASHPTE,
+#endif
+               .set    = "hpte",
+               .clear  = "    ",
+       }, {
+#ifndef CONFIG_PPC_STD_MMU_64
+               .mask   = _PAGE_GUARDED,
+               .val    = _PAGE_GUARDED,
+               .set    = "guarded",
+               .clear  = "       ",
+       }, {
+#endif
+               .mask   = _PAGE_DIRTY,
+               .val    = _PAGE_DIRTY,
+               .set    = "dirty",
+               .clear  = "     ",
+       }, {
+               .mask   = _PAGE_ACCESSED,
+               .val    = _PAGE_ACCESSED,
+               .set    = "accessed",
+               .clear  = "        ",
+       }, {
+#ifndef CONFIG_PPC_STD_MMU_64
+               .mask   = _PAGE_WRITETHRU,
+               .val    = _PAGE_WRITETHRU,
+               .set    = "write through",
+               .clear  = "             ",
+       }, {
+#endif
+               .mask   = _PAGE_NO_CACHE,
+               .val    = _PAGE_NO_CACHE,
+               .set    = "no cache",
+               .clear  = "        ",
+       }, {
+               .mask   = H_PAGE_BUSY,
+               .val    = H_PAGE_BUSY,
+               .set    = "busy",
+       }, {
+#ifdef CONFIG_PPC_64K_PAGES
+               .mask   = H_PAGE_COMBO,
+               .val    = H_PAGE_COMBO,
+               .set    = "combo",
+       }, {
+               .mask   = H_PAGE_4K_PFN,
+               .val    = H_PAGE_4K_PFN,
+               .set    = "4K_pfn",
+       }, {
+#endif
+               .mask   = H_PAGE_F_GIX,
+               .val    = H_PAGE_F_GIX,
+               .set    = "f_gix",
+               .is_val = true,
+               .shift  = H_PAGE_F_GIX_SHIFT,
+       }, {
+               .mask   = H_PAGE_F_SECOND,
+               .val    = H_PAGE_F_SECOND,
+               .set    = "f_second",
+       }, {
+               .mask   = _PAGE_SPECIAL,
+               .val    = _PAGE_SPECIAL,
+               .set    = "special",
+       }
+};
+
+struct pgtable_level {
+       const struct flag_info *flag;
+       size_t num;
+       u64 mask;
+};
+
+static struct pgtable_level pg_level[] = {
+       {
+       }, { /* pgd */
+               .flag   = flag_array,
+               .num    = ARRAY_SIZE(flag_array),
+       }, { /* pud */
+               .flag   = flag_array,
+               .num    = ARRAY_SIZE(flag_array),
+       }, { /* pmd */
+               .flag   = flag_array,
+               .num    = ARRAY_SIZE(flag_array),
+       }, { /* pte */
+               .flag   = flag_array,
+               .num    = ARRAY_SIZE(flag_array),
+       },
+};
+
+static void dump_flag_info(struct pg_state *st, const struct flag_info
+               *flag, u64 pte, int num)
+{
+       unsigned int i;
+
+       for (i = 0; i < num; i++, flag++) {
+               const char *s = NULL;
+               u64 val;
+
+               /* flag not defined so don't check it */
+               if (flag->mask == 0)
+                       continue;
+               /* Some 'flags' are actually values */
+               if (flag->is_val) {
+                       val = pte & flag->val;
+                       if (flag->shift)
+                               val = val >> flag->shift;
+                       seq_printf(st->seq, "  %s:%llx", flag->set, val);
+               } else {
+                       if ((pte & flag->mask) == flag->val)
+                               s = flag->set;
+                       else
+                               s = flag->clear;
+                       if (s)
+                               seq_printf(st->seq, "  %s", s);
+               }
+               st->current_flags &= ~flag->mask;
+       }
+       if (st->current_flags != 0)
+               seq_printf(st->seq, "  unknown flags:%llx", st->current_flags);
+}
+
+static void dump_addr(struct pg_state *st, unsigned long addr)
+{
+       static const char units[] = "KMGTPE";
+       const char *unit = units;
+       unsigned long delta;
+
+       seq_printf(st->seq, "0x%016lx-0x%016lx   ", st->start_address, addr-1);
+       delta = (addr - st->start_address) >> 10;
+       /* Work out what appropriate unit to use */
+       while (!(delta & 1023) && unit[1]) {
+               delta >>= 10;
+               unit++;
+       }
+       seq_printf(st->seq, "%9lu%c", delta, *unit);
+
+}
+
+static void note_page(struct pg_state *st, unsigned long addr,
+              unsigned int level, u64 val)
+{
+       u64 flag = val & pg_level[level].mask;
+       /* At first no level is set */
+       if (!st->level) {
+               st->level = level;
+               st->current_flags = flag;
+               st->start_address = addr;
+               seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
+       /*
+        * Dump the section of virtual memory when:
+        *   - the PTE flags from one entry to the next differs.
+        *   - we change levels in the tree.
+        *   - the address is in a different section of memory and is thus
+        *   used for a different purpose, regardless of the flags.
+        */
+       } else if (flag != st->current_flags || level != st->level ||
+                  addr >= st->marker[1].start_address) {
+
+               /* Check the PTE flags */
+               if (st->current_flags) {
+                       dump_addr(st, addr);
+
+                       /* Dump all the flags */
+                       if (pg_level[st->level].flag)
+                               dump_flag_info(st, pg_level[st->level].flag,
+                                         st->current_flags,
+                                         pg_level[st->level].num);
+
+                       seq_puts(st->seq, "\n");
+               }
+
+               /*
+                * Address indicates we have passed the end of the
+                * current section of virtual memory
+                */
+               while (addr >= st->marker[1].start_address) {
+                       st->marker++;
+                       seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
+               }
+               st->start_address = addr;
+               st->current_flags = flag;
+               st->level = level;
+       }
+}
+
+static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
+{
+       pte_t *pte = pte_offset_kernel(pmd, 0);
+       unsigned long addr;
+       unsigned int i;
+
+       for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
+               addr = start + i * PAGE_SIZE;
+               note_page(st, addr, 4, pte_val(*pte));
+
+       }
+}
+
+static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
+{
+       pmd_t *pmd = pmd_offset(pud, 0);
+       unsigned long addr;
+       unsigned int i;
+
+       for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
+               addr = start + i * PMD_SIZE;
+               if (!pmd_none(*pmd))
+                       /* pmd exists */
+                       walk_pte(st, pmd, addr);
+               else
+                       note_page(st, addr, 3, pmd_val(*pmd));
+       }
+}
+
+static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start)
+{
+       pud_t *pud = pud_offset(pgd, 0);
+       unsigned long addr;
+       unsigned int i;
+
+       for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
+               addr = start + i * PUD_SIZE;
+               if (!pud_none(*pud))
+                       /* pud exists */
+                       walk_pmd(st, pud, addr);
+               else
+                       note_page(st, addr, 2, pud_val(*pud));
+       }
+}
+
+static void walk_pagetables(struct pg_state *st)
+{
+       pgd_t *pgd = pgd_offset_k(0UL);
+       unsigned int i;
+       unsigned long addr;
+
+       /*
+        * Traverse the linux pagetable structure and dump pages that are in
+        * the hash pagetable.
+        */
+       for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
+               addr = KERN_VIRT_START + i * PGDIR_SIZE;
+               if (!pgd_none(*pgd))
+                       /* pgd exists */
+                       walk_pud(st, pgd, addr);
+               else
+                       note_page(st, addr, 1, pgd_val(*pgd));
+       }
+}
+
+static void populate_markers(void)
+{
+       address_markers[0].start_address = PAGE_OFFSET;
+       address_markers[1].start_address = VMALLOC_START;
+       address_markers[2].start_address = VMALLOC_END;
+       address_markers[3].start_address = ISA_IO_BASE;
+       address_markers[4].start_address = ISA_IO_END;
+       address_markers[5].start_address = PHB_IO_BASE;
+       address_markers[6].start_address = PHB_IO_END;
+       address_markers[7].start_address = IOREMAP_BASE;
+       address_markers[8].start_address = IOREMAP_END;
+#ifdef CONFIG_PPC_STD_MMU_64
+       address_markers[9].start_address =  H_VMEMMAP_BASE;
+#else
+       address_markers[9].start_address =  VMEMMAP_BASE;
+#endif
+}
+
+static int ptdump_show(struct seq_file *m, void *v)
+{
+       struct pg_state st = {
+               .seq = m,
+               .start_address = KERN_VIRT_START,
+               .marker = address_markers,
+       };
+       /* Traverse kernel page tables */
+       walk_pagetables(&st);
+       note_page(&st, 0, 0, 0);
+       return 0;
+}
+
+
+static int ptdump_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, ptdump_show, NULL);
+}
+
+static const struct file_operations ptdump_fops = {
+       .open           = ptdump_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static void build_pgtable_complete_mask(void)
+{
+       unsigned int i, j;
+
+       for (i = 0; i < ARRAY_SIZE(pg_level); i++)
+               if (pg_level[i].flag)
+                       for (j = 0; j < pg_level[i].num; j++)
+                               pg_level[i].mask |= pg_level[i].flag[j].mask;
+}
+
+static int ptdump_init(void)
+{
+       struct dentry *debugfs_file;
+
+       populate_markers();
+       build_pgtable_complete_mask();
+       debugfs_file = debugfs_create_file("kernel_pagetables", 0400, NULL,
+                       NULL, &ptdump_fops);
+       return debugfs_file ? 0 : -ENOMEM;
+}
+device_initcall(ptdump_init);