x86: add code to dump the (kernel) page tables for visual inspection by kernel developers
authorArjan van de Ven <arjan@linux.intel.com>
Thu, 17 Apr 2008 15:40:45 +0000 (17:40 +0200)
committerIngo Molnar <mingo@elte.hu>
Thu, 17 Apr 2008 15:40:45 +0000 (17:40 +0200)
This patch adds code to the kernel to have an (optional)
/proc/kernel_page_tables debug file that basically dumps the kernel
pagetables; this allows us kernel developers to verify that nothing fishy is
going on and that the various mappings are set up correctly. This was quite
useful in finding various change_page_attr() bugs, and is very likely to be
useful in the future as well.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: mingo@elte.hu
Cc: tglx@tglx.de
Cc: hpa@zytor.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
arch/x86/Kconfig.debug
arch/x86/mm/Makefile
arch/x86/mm/dump_pagetables.c [new file with mode: 0644]

index 702eb39901ca3ee987f0adeb81f7d38567ba8c93..cb7002eca8875000b69e2c9165a8f9187b2942f6 100644 (file)
@@ -54,6 +54,18 @@ config DEBUG_PER_CPU_MAPS
 
          Say N if unsure.
 
+config X86_PTDUMP
+       bool "Export kernel pagetable layout to userspace via debugfs"
+       depends on X86_64
+       select DEBUG_FS
+       help
+         Say Y here if you want to show the kernel pagetable layout in a
+         debugfs file. This information is only useful for kernel developers
+         who are working in architecture specific areas of the kernel.
+         It is probably not a good idea to enable this feature in a production
+         kernel.
+         If in doubt, say "N"
+
 config DEBUG_RODATA
        bool "Write protect kernel read-only data structures"
        default y
index 8e81660604bc7f5e15b36ffa347477f2f5708783..28632f42ca66173c50e2e528a7e3f35ac7a09818 100644 (file)
@@ -12,4 +12,5 @@ else
 obj-$(CONFIG_NUMA)             += numa_64.o
 obj-$(CONFIG_K8_NUMA)          += k8topology_64.o
 obj-$(CONFIG_ACPI_NUMA)                += srat_64.o
+obj-$(CONFIG_X86_PTDUMP)       += dump_pagetables.o
 endif
diff --git a/arch/x86/mm/dump_pagetables.c b/arch/x86/mm/dump_pagetables.c
new file mode 100644 (file)
index 0000000..5e7f643
--- /dev/null
@@ -0,0 +1,301 @@
+/*
+ * Debug helper to dump the current kernel pagetables of the system
+ * so that we can see what the various memory ranges are set to.
+ *
+ * (C) Copyright 2008 Intel Corporation
+ *
+ * Author: Arjan van de Ven <arjan@linux.intel.com>
+ *
+ * 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/module.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
+
+#include <asm/pgtable.h>
+
+/*
+ * The dumper groups pagetable entries of the same type into one, and for
+ * that it needs to keep some state when walking, and flush this state
+ * when a "break" in the continuity is found.
+ */
+struct pg_state {
+       int level;
+       pgprot_t current_prot;
+       unsigned long start_address;
+       unsigned long current_address;
+       int printed_vmalloc;
+       int printed_modules;
+       int printed_vmemmap;
+       int printed_highmap;
+};
+
+/* Multipliers for offsets within the PTEs */
+#define LEVEL_4_MULT (PAGE_SIZE)
+#define LEVEL_3_MULT (512UL * LEVEL_4_MULT)
+#define LEVEL_2_MULT (512UL * LEVEL_3_MULT)
+#define LEVEL_1_MULT (512UL * LEVEL_2_MULT)
+
+
+/*
+ * Print a readable form of a pgprot_t to the seq_file
+ */
+static void printk_prot(struct seq_file *m, pgprot_t prot, int level)
+{
+       unsigned long pr = pgprot_val(prot);
+
+       if (pr & _PAGE_USER)
+               seq_printf(m, "USR ");
+       else
+               seq_printf(m, "    ");
+       if (pr & _PAGE_RW)
+               seq_printf(m, "RW ");
+       else
+               seq_printf(m, "ro ");
+       if (pr & _PAGE_PWT)
+               seq_printf(m, "PWT ");
+       else
+               seq_printf(m, "    ");
+       if (pr & _PAGE_PCD)
+               seq_printf(m, "PCD ");
+       else
+               seq_printf(m, "    ");
+
+       /* Bit 9 has a different meaning on level 3 vs 4 */
+       if (level <= 3) {
+               if (pr & _PAGE_PSE)
+                       seq_printf(m, "PSE ");
+               else
+                       seq_printf(m, "    ");
+       } else {
+               if (pr & _PAGE_PAT)
+                       seq_printf(m, "pat ");
+               else
+                       seq_printf(m, "    ");
+       }
+       if (pr & _PAGE_GLOBAL)
+               seq_printf(m, "GLB ");
+       else
+               seq_printf(m, "    ");
+       if (pr & _PAGE_NX)
+               seq_printf(m, "NX ");
+       else
+               seq_printf(m, "x  ");
+}
+
+/*
+ * Sign-extend the 48 bit address to 64 bit
+ */
+static unsigned long sign_extend(unsigned long u)
+{
+       if (u>>47)
+               u = u | (0xffffUL << 48);
+       return u;
+}
+
+/*
+ * This function gets called on a break in a continuous series
+ * of PTE entries; the next one is different so we need to
+ * print what we collected so far.
+ */
+static void note_page(struct seq_file *m, struct pg_state *st,
+                                       pgprot_t new_prot, int level)
+{
+       unsigned long prot, cur;
+
+       /*
+        * If we have a "break" in the series, we need to flush the state that
+        * we have now. "break" is either changing perms or a different level.
+        */
+       prot = pgprot_val(new_prot) & ~(PTE_MASK);
+       cur = pgprot_val(st->current_prot) & ~(PTE_MASK);
+
+       if ((prot != cur || level != st->level) &&
+                               st->current_address != st->start_address) {
+               char unit = 'K';
+               unsigned long delta;
+
+               /*
+                * We print markers for special areas of address space,
+                * such as the start of vmalloc space etc.
+                * This helps in the interpretation.
+                */
+               if (!st->printed_vmalloc &&
+                               st->start_address >= VMALLOC_START) {
+                       seq_printf(m, "---[ VMALLOC SPACE ]---\n");
+                       st->printed_vmalloc = 1;
+               }
+               if (!st->printed_modules &&
+                               st->start_address >= MODULES_VADDR) {
+                       seq_printf(m, "---[ MODULES SPACE ]---\n");
+                       st->printed_modules = 1;
+               }
+               if (st->printed_modules < 2 &&
+                               st->start_address >= MODULES_END) {
+                       seq_printf(m, "---[ END MODULES SPACE ]---\n");
+                       st->printed_modules = 2;
+               }
+               if (!st->printed_vmemmap &&
+                               st->start_address >= VMEMMAP_START) {
+                       seq_printf(m, "---[ VMMEMMAP SPACE ]---\n");
+                       st->printed_vmemmap = 1;
+               }
+               if (!st->printed_highmap &&
+                               st->start_address >= __START_KERNEL_map) {
+                       seq_printf(m, "---[ HIGH KERNEL MAPPING ]---\n");
+                       st->printed_highmap = 1;
+               }
+
+               /*
+                * Now print the actual finished series
+                */
+               seq_printf(m, "[ %016lx -  %016lx   ",
+                               st->start_address, st->current_address);
+
+               delta = (st->current_address - st->start_address) >> 10;
+               if ((delta & 1023) == 0) {
+                       delta = delta >> 10;
+                       unit = 'M';
+               }
+               if (pgprot_val(st->current_prot)) {
+                       seq_printf(m, "Size %9lu%cb ", delta, unit);
+                       printk_prot(m, st->current_prot, st->level);
+                       seq_printf(m, "L%i]\n", st->level);
+               } else {
+                       /* don't print protections on non-present memory */
+                       seq_printf(m, "%14lu%cb", delta, unit);
+                       seq_printf(m, "                           L%i]\n",
+                                       st->level);
+               }
+               st->start_address = st->current_address;
+               st->current_prot = new_prot;
+               st->level = level;
+       };
+}
+
+static void walk_level_4(struct seq_file *m, struct pg_state *st, pmd_t addr,
+                                                       unsigned long P)
+{
+       int i;
+       pte_t *start;
+
+       start = (pte_t *) pmd_page_vaddr(addr);
+       for (i = 0; i < PTRS_PER_PTE; i++) {
+               pgprot_t prot = pte_pgprot(*start);
+
+               st->current_address = sign_extend(P + i * LEVEL_4_MULT);
+               note_page(m, st, prot, 4);
+               start++;
+       }
+}
+
+
+static void walk_level_3(struct seq_file *m, struct pg_state *st, pud_t addr,
+                                                       unsigned long P)
+{
+       int i;
+       pmd_t *start;
+
+       start = (pmd_t *) pud_page_vaddr(addr);
+       for (i = 0; i < PTRS_PER_PMD; i++) {
+               st->current_address = sign_extend(P + i * LEVEL_3_MULT);
+               if (!pmd_none(*start)) {
+                       unsigned long prot;
+
+                       prot = pmd_val(*start) & ~(PTE_MASK);
+                       /* Deal with 2Mb pages */
+                       if (pmd_large(*start))
+                               note_page(m, st, __pgprot(prot), 3);
+                       else
+                               walk_level_4(m, st, *start,
+                                                       P + i * LEVEL_3_MULT);
+               } else
+                       note_page(m, st, __pgprot(0), 3);
+               start++;
+       }
+}
+
+
+static void walk_level_2(struct seq_file *m, struct pg_state *st, pgd_t addr,
+                                                       unsigned long P)
+{
+       int i;
+       pud_t *start;
+
+       start = (pud_t *) pgd_page_vaddr(addr);
+
+       for (i = 0; i < PTRS_PER_PUD; i++) {
+               if (!pud_none(*start)) {
+                       unsigned long prot;
+
+                       prot = pud_val(*start) & ~(PTE_MASK);
+                       /* Deal with 1Gb pages */
+                       if (pud_large(*start))
+                               note_page(m, st, __pgprot(prot), 2);
+                       else
+                               walk_level_3(m, st, *start,
+                                       P + i * LEVEL_2_MULT);
+               } else
+                       note_page(m, st, __pgprot(0), 2);
+
+               start++;
+       }
+}
+
+static void walk_level_1(struct seq_file *m)
+{
+       pgd_t *start = (pgd_t *) &init_level4_pgt;
+       int i;
+       struct pg_state st;
+
+       memset(&st, 0, sizeof(st));
+       st.level = 1;
+
+       for (i = 0; i < PTRS_PER_PGD; i++) {
+               if (!pgd_none(*start))
+                       walk_level_2(m, &st, *start, i * LEVEL_1_MULT);
+               else
+                       note_page(m, &st, __pgprot(0), 1);
+               start++;
+       }
+}
+
+static int ptdump_show(struct seq_file *m, void *v)
+{
+       seq_puts(m, "Kernel pagetable dump\n");
+       walk_level_1(m);
+       return 0;
+}
+
+static int ptdump_open(struct inode *inode, struct file *filp)
+{
+       return single_open(filp, ptdump_show, NULL);
+}
+
+static const struct file_operations ptdump_fops = {
+       .open           = ptdump_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+int pt_dump_init(void)
+{
+       struct dentry *pe;
+
+       pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
+                                &ptdump_fops);
+       if (!pe)
+               return -ENOMEM;
+
+       return 0;
+}
+
+__initcall(pt_dump_init);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
+MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");