From mboxrd@z Thu Jan 1 00:00:00 1970 From: Steve Capper Subject: Re: ARM: add support to dump the kernel page tables Date: Thu, 24 Oct 2013 11:51:44 +0100 Message-ID: <20131024105143.GA24288@linaro.org> References: <20131024071600.GC16735@n2100.arm.linux.org.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from mail-we0-f178.google.com ([74.125.82.178]:52591 "EHLO mail-we0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754003Ab3JXKvv (ORCPT ); Thu, 24 Oct 2013 06:51:51 -0400 Received: by mail-we0-f178.google.com with SMTP id q59so2086437wes.23 for ; Thu, 24 Oct 2013 03:51:50 -0700 (PDT) Content-Disposition: inline In-Reply-To: <20131024071600.GC16735@n2100.arm.linux.org.uk> Sender: linux-arch-owner@vger.kernel.org List-ID: To: Russell King - ARM Linux Cc: linux-arm-kernel@lists.infradead.org, linux-arch@vger.kernel.org On Thu, Oct 24, 2013 at 08:16:00AM +0100, Russell King - ARM Linux wrot= e: > This is very similar to x86. I'm just throwing the patch out if peop= le > wish to use this to look at things; I'm going to be working on mergin= g > it with the x86 version, and hopefully we can have the bulk of this > support provided in a generic way such that architectures just need t= o > define some bitfield data, macros and region data. Hi Russell, I've given this a quick go on an Arndale with LPAE, and have a few suggestions below. >=20 > 8<=3D=3D=3D > From: Russell King > ARM: add support to dump the kernel page tables >=20 > This patch allows the kernel page tables to be dumped via a debugfs f= ile, > allowing kernel developers to check the layout of the kernel page tab= les > and the verify the various permissions and type settings. >=20 > Signed-off-by: Russell King > --- > arch/arm/Kconfig.debug | 12 ++ > arch/arm/include/asm/pgtable-2level.h | 1 + > arch/arm/include/asm/pgtable-3level.h | 1 + > arch/arm/mm/Makefile | 1 + > arch/arm/mm/dump.c | 328 +++++++++++++++++++++++= ++++++++++ > 5 files changed, 343 insertions(+), 0 deletions(-) > create mode 100644 arch/arm/mm/dump.c >=20 > diff --git a/arch/arm/Kconfig.debug b/arch/arm/Kconfig.debug > index 583f4a0..261cc75 100644 > --- a/arch/arm/Kconfig.debug > +++ b/arch/arm/Kconfig.debug > @@ -2,6 +2,18 @@ menu "Kernel hacking" > =20 > source "lib/Kconfig.debug" > =20 > +config ARM_PTDUMP > + bool "Export kernel pagetable layout ot userspace via debugfs" > + depends on DEBUG_KERNEL > + 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 develope= rs > + who are working in architecture specific areas of the kernel. > + It is probably not a good idea to enable this feature in a produc= tion > + kernel. > + If in doubt, say "N" > + > config STRICT_DEVMEM > bool "Filter access to /dev/mem" > depends on MMU > diff --git a/arch/arm/include/asm/pgtable-2level.h b/arch/arm/include= /asm/pgtable-2level.h > index f97ee02..b082d00 100644 > --- a/arch/arm/include/asm/pgtable-2level.h > +++ b/arch/arm/include/asm/pgtable-2level.h > @@ -160,6 +160,7 @@ static inline pmd_t *pmd_offset(pud_t *pud, unsig= ned long addr) > return (pmd_t *)pud; > } > =20 > +#define pmd_large(pmd) (pmd_val(pmd) & 2) > #define pmd_bad(pmd) (pmd_val(pmd) & 2) > =20 > #define copy_pmd(pmdpd,pmdps) \ > diff --git a/arch/arm/include/asm/pgtable-3level.h b/arch/arm/include= /asm/pgtable-3level.h > index 5689c18..d7682cd 100644 > --- a/arch/arm/include/asm/pgtable-3level.h > +++ b/arch/arm/include/asm/pgtable-3level.h > @@ -140,6 +140,7 @@ > PMD_TYPE_TABLE) > #define pmd_sect(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) =3D=3D \ > PMD_TYPE_SECT) > +#define pmd_large(pmd) pmd_sect(pmd) Could we please instead do something like: #define pmd_large(pmd) (pmd_val(pmd) && !(pmd_val(pmd) & PMD_TA= BLE_BIT)) This matches the logic used in the huge pages patches (it picks up the PROT_NONE case) that would be missed above. > =20 > #define pud_clear(pudp) \ > do { \ > diff --git a/arch/arm/mm/Makefile b/arch/arm/mm/Makefile > index ecfe6e5..7f39ce2 100644 > --- a/arch/arm/mm/Makefile > +++ b/arch/arm/mm/Makefile > @@ -12,6 +12,7 @@ ifneq ($(CONFIG_MMU),y) > obj-y +=3D nommu.o > endif > =20 > +obj-$(CONFIG_ARM_PTDUMP) +=3D dump.o > obj-$(CONFIG_MODULES) +=3D proc-syms.o > =20 > obj-$(CONFIG_ALIGNMENT_TRAP) +=3D alignment.o > diff --git a/arch/arm/mm/dump.c b/arch/arm/mm/dump.c > new file mode 100644 > index 0000000..4979d4c > --- /dev/null > +++ b/arch/arm/mm/dump.c > @@ -0,0 +1,328 @@ > +/* > + * Debug helper to dump the current kernel pagetables of the system > + * so that we can see what the various memory ranges are set to. > + * > + * Derived from x86 implementation: > + * (C) Copyright 2008 Intel Corporation > + * > + * Author: 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 > +#include > +#include > +#include > + > +#include > +#include > + > +struct addr_marker { > + unsigned long start_address; > + const char *name; > +}; > + > +static struct addr_marker address_markers[] =3D { > + { MODULES_VADDR, "Modules" }, > + { PAGE_OFFSET, "Kernel Mapping" }, > + { 0, "vmalloc() Area" }, > + { VMALLOC_END, "vmalloc() End" }, > + { FIXADDR_START, "Fixmap Area" }, > + { CONFIG_VECTORS_BASE, "Vectors" }, > + { CONFIG_VECTORS_BASE + PAGE_SIZE * 2, "Vectors End" }, > + { -1, NULL }, > +}; > + > +struct pg_state { > + struct seq_file *seq; > + const struct addr_marker *marker; > + unsigned long start_address; > + unsigned level; > + u64 current_prot; > +}; > + > +struct prot_bits { > + u64 mask; > + u64 val; > + const char *set; > + const char *clear; > +}; > + > +static const struct prot_bits pte_bits[] =3D { > + { > + .mask =3D L_PTE_USER, > + .val =3D L_PTE_USER, > + .set =3D "USR", > + .clear =3D " ", > + }, { > + .mask =3D L_PTE_RDONLY, > + .val =3D L_PTE_RDONLY, > + .set =3D "ro", > + .clear =3D "RW", > + }, { > + .mask =3D L_PTE_XN, > + .val =3D L_PTE_XN, > + .set =3D "NX", > + .clear =3D "x ", > + }, { > + .mask =3D L_PTE_SHARED, > + .val =3D L_PTE_SHARED, > + .set =3D "SHD", > + .clear =3D " ", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_UNCACHED, > + .set =3D "SO/UNCACHED", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_BUFFERABLE, > + .set =3D "MEM/BUFFERABLE/WC", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_WRITETHROUGH, > + .set =3D "MEM/CACHED/WT", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_WRITEBACK, > + .set =3D "MEM/CACHED/WBRA", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_MINICACHE, > + .set =3D "MEM/MINICACHE", > + }, { This throws the following compile error when I have LPAE enabled: arch/arm/mm/dump.c:93:10: error: =E2=80=98L_PTE_MT_MINICACHE=E2=80=99 u= ndeclared here (not in a function) > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_WRITEALLOC, > + .set =3D "MEM/CACHED/WBWA", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_DEV_SHARED, > + .set =3D "DEV/SHARED", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_DEV_NONSHARED, > + .set =3D "DEV/NONSHARED", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_DEV_WC, > + .set =3D "DEV/WC", > + }, { > + .mask =3D L_PTE_MT_MASK, > + .val =3D L_PTE_MT_DEV_CACHED, > + .set =3D "DEV/CACHED", > + }, > +}; > + > +static const struct prot_bits section_bits[] =3D { > + /* These are approximate */ > + { > + .mask =3D PMD_SECT_AP_READ | PMD_SECT_AP_WRITE, > + .val =3D 0, > + .set =3D " ro", > + }, { > + .mask =3D PMD_SECT_AP_READ | PMD_SECT_AP_WRITE, > + .val =3D PMD_SECT_AP_WRITE, > + .set =3D " RW", > + }, { > + .mask =3D PMD_SECT_AP_READ | PMD_SECT_AP_WRITE, > + .val =3D PMD_SECT_AP_READ, > + .set =3D "USR RO", > + }, { > + .mask =3D PMD_SECT_AP_READ | PMD_SECT_AP_WRITE, > + .val =3D PMD_SECT_AP_READ | PMD_SECT_AP_WRITE, > + .set =3D "USR RW", > + }, { > + .mask =3D PMD_SECT_XN, > + .val =3D PMD_SECT_XN, > + .set =3D "NX", > + .clear =3D "x ", > + }, { > + .mask =3D PMD_SECT_S, > + .val =3D PMD_SECT_S, > + .set =3D "SHD", > + .clear =3D " ", > + }, > +}; > + > +struct pg_level { > + const struct prot_bits *bits; > + size_t num; > + u64 mask; > +}; > + > +static struct pg_level pg_level[] =3D { > + { > + }, { /* pgd */ > + }, { /* pud */ > + }, { /* pmd */ > + .bits =3D section_bits, > + .num =3D ARRAY_SIZE(section_bits), > + }, { /* pte */ > + .bits =3D pte_bits, > + .num =3D ARRAY_SIZE(pte_bits), > + }, > +}; > + > +static void dump_prot(struct pg_state *st, const struct prot_bits *b= its, size_t num) > +{ > + unsigned i; > + > + for (i =3D 0; i < num; i++, bits++) { > + const char *s; > + > + if ((st->current_prot & bits->mask) =3D=3D bits->val) > + s =3D bits->set; > + else > + s =3D bits->clear; > + > + if (s) > + seq_printf(st->seq, " %s", s); > + } > +} > + > +static void note_page(struct pg_state *st, unsigned long addr, unsig= ned level, u64 val) > +{ > + static const char units[] =3D "KMGTPE"; > + u64 prot =3D val & pg_level[level].mask; > + > + if (addr < USER_PGTABLES_CEILING) > + return; > + > + if (!st->level) { > + st->level =3D level; > + st->current_prot =3D prot; > + seq_printf(st->seq, "---[ %s ]---\n", st->marker->name); > + } else if (prot !=3D st->current_prot || level !=3D st->level || > + addr >=3D st->marker[1].start_address) { > + const char *unit =3D units; > + unsigned long delta; > + > + if (st->current_prot) { > + seq_printf(st->seq, "0x%08lx-0x%08lx ", > + st->start_address, addr); > + > + delta =3D (addr - st->start_address) >> 10; > + while (!(delta & 1023) && unit[1]) { > + delta >>=3D 10; > + unit++; > + } > + seq_printf(st->seq, "%9lu%c", delta, *unit); > + if (pg_level[st->level].bits) > + dump_prot(st, pg_level[st->level].bits, pg_level[st->level].num)= ; > + seq_printf(st->seq, "\n"); > + } > + > + if (addr >=3D st->marker[1].start_address) { > + st->marker++; > + seq_printf(st->seq, "---[ %s ]---\n", st->marker->name); > + } > + st->start_address =3D addr; > + st->current_prot =3D prot; > + st->level =3D level; > + } > +} > + > +static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long = start) > +{ > + pte_t *pte =3D pte_offset_kernel(pmd, 0); > + unsigned long addr; > + unsigned i; > + > + for (i =3D 0; i < PTRS_PER_PTE; i++, pte++) { > + addr =3D 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 =3D pmd_offset(pud, 0); > + unsigned long addr; > + unsigned i; > + > + for (i =3D 0; i < PTRS_PER_PMD; i++, pmd++) { > + addr =3D start + i * PMD_SIZE; > + if (pmd_none(*pmd) || pmd_large(*pmd) || !pmd_present(*pmd)) > + note_page(st, addr, 3, pmd_val(*pmd)); > + else > + walk_pte(st, pmd, addr); > + } > +} > + > +static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long = start) > +{ > + pud_t *pud =3D pud_offset(pgd, 0); > + unsigned long addr; > + unsigned i; > + > + for (i =3D 0; i < PTRS_PER_PUD; i++, pud++) { > + addr =3D start + i * PUD_SIZE; > + if (!pud_none(*pud)) { > + walk_pmd(st, pud, addr); > + } else { > + note_page(st, addr, 2, pud_val(*pud)); > + } > + } > +} > + > +static void walk_pgd(struct seq_file *m) > +{ > + pgd_t *pgd =3D swapper_pg_dir; > + struct pg_state st; > + unsigned long addr; > + unsigned i; > + > + memset(&st, 0, sizeof(st)); > + st.seq =3D m; > + st.marker =3D address_markers; > + > + for (i =3D USER_PGTABLES_CEILING / PGDIR_SIZE; > + i < PTRS_PER_PGD; i++, pgd++) { > + addr =3D i * PGDIR_SIZE; > + if (!pgd_none(*pgd)) { > + walk_pud(&st, pgd, addr); > + } else { > + note_page(&st, addr, 1, pgd_val(*pgd)); > + } > + } > + > + note_page(&st, 0, 0, 0); > +} The logic here needs adjusting for LPAE. Without LPAE, USER_PGTABLES_CEILING is equal to zero and i at 0. With LPAE enabled, USER_PGTABLES_CEILING / PGDIR_SIZE =3D=3D 2 for me, = but pgd refers to the first pgd (not the third). Thus we walk the pgds we don't want to walk and get empty output. The following worked for me: static void walk_pgd(struct seq_file *m) { pgd_t *pgd =3D swapper_pg_dir; struct pg_state st; unsigned long addr; unsigned i, pgdoff =3D USER_PGTABLES_CEILING / PGDIR_SIZE; memset(&st, 0, sizeof(st)); st.seq =3D m; st.marker =3D address_markers; pgd +=3D pgdoff; for (i =3D pgdoff; i < PTRS_PER_PGD; i++, pgd++) { addr =3D i * PGDIR_SIZE; if (!pgd_none(*pgd)) { walk_pud(&st, pgd, addr); } else { note_page(&st, addr, 1, pgd_val(*pgd)); } } note_page(&st, 0, 0, 0); } But pgdoff is a terrible variable name :-). > + > +static int ptdump_show(struct seq_file *m, void *v) > +{ > + walk_pgd(m); > + 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 =3D { > + .open =3D ptdump_open, > + .read =3D seq_read, > + .llseek =3D seq_lseek, > + .release =3D single_release, > +}; > + > +static int ptdump_init(void) > +{ > + struct dentry *pe; > + unsigned i, j; > + > + for (i =3D 0; i < ARRAY_SIZE(pg_level); i++) > + if (pg_level[i].bits) > + for (j =3D 0; j < pg_level[i].num; j++) > + pg_level[i].mask |=3D pg_level[i].bits[j].mask; > + > + address_markers[2].start_address =3D VMALLOC_START; > + > + pe =3D debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, > + &ptdump_fops); > + return pe ? 0 : -ENOMEM; > +} > +__initcall(ptdump_init); > --=20 > 1.7.4.4 Cheers, --=20 Steve