Hello,
* Bernhard Walle <bwalle at suse.de> [2007-05-14 23:49]:
>
> I'm working on support for the 4 layer pagetable.
Here's my first attempt. It works with SLES 10, SP1 on a Tiger4
machine with 16 GiB of memory. (I had still problems with a big
SGI machine when creating the bitmap. I'm investigating this, too.)
The problem is not to implement the 4 layer page table, but to
*detect* it. crash uses the built-in configuration data in the kernel
image, and that's what I used in my patch. If you have a better and
still reliable method, I'm open to suggestions. :-)
Also, I only use the page table translation when the memory is
vmalloc'd. That's the same way crash does it and that should be
well-tested.
The patch requires your two patches. Please give me your feedback!
And I hope I didn't accidentally break the 3 layer page table.
Thanks,
Bernhard
---
ia64.c | 148 +++++++++++++++++++++++++++++++++++++++++++++---
makedumpfile.c | 174 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
makedumpfile.h | 43 +++++++++++++-
3 files changed, 354 insertions(+), 11 deletions(-)
--- a/ia64.c
+++ b/ia64.c
@@ -3,6 +3,9 @@
*
* Copyright (C) 2006 NEC Corporation
*
+ * Some parts are taken and adapted from the crash-utility,
+ * (c) by RedHat Inc.
+ *
* 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; either version 2 of the License, or
@@ -17,6 +20,18 @@
#include "makedumpfile.h"
+
+/*
+ * vmalloc() starting address is either the traditional 0xa000000000000000 or
+ * bumped up in 2.6 to 0xa000000200000000.
+ */
+int
+is_vmalloc_addr_ia64(struct DumpInfo *info, unsigned long vaddr)
+{
+ return ((vaddr >= info->vmalloc_start) &&
+ (vaddr < (ulong)KERNEL_UNCACHED_BASE));
+}
+
int
get_phys_base_ia64(struct DumpInfo *info)
{
@@ -51,18 +66,80 @@ get_machdep_info_ia64(struct DumpInfo *i
}
unsigned long
-ia64_vtop(struct DumpInfo *info, unsigned long long vaddr)
+ia64_vtop3(struct DumpInfo *info, unsigned long long vaddr)
{
unsigned long paddr = 0, temp, page_dir, pgd_pte, page_middle, pmd_pte;
unsigned long page_table, pte;
- if (VADDR_REGION(vaddr) != KERNEL_VMALLOC_REGION) {
- ERRMSG("vaddr(%llx) is not KERNEL_VMALLOC_REGION.\n", vaddr);
+
+ /*
+ * Translate a virtual address to a physical address
+ * by using Layer 3 paging.
+ */
+ if (SYMBOL(swapper_pg_dir) == NOT_FOUND_SYMBOL) {
+ ERRMSG("Can't get the symbol of swapper_pg_dir.\n");
return paddr;
}
- paddr = vaddr_to_paddr(info, vaddr);
- if (paddr)
+
+ /*
+ * Get PGD
+ */
+ temp = vaddr & MASK_PGD_3L;
+ temp = temp >> (PGDIR_SHIFT_3L - 3);
+ page_dir = SYMBOL(swapper_pg_dir) + temp;
+ if (!readmem(info, page_dir, &pgd_pte, sizeof pgd_pte)) {
+ ERRMSG("Can't get pgd_pte (page_dir:%lx).\n", page_dir);
return paddr;
+ }
+
+ /*
+ * Get PMD
+ */
+ temp = vaddr & MASK_PMD;
+ temp = temp >> (PMD_SHIFT - 3);
+ page_middle = pgd_pte + temp;
+ /*
+ * Convert physical address to virtual address
+ */
+ page_middle = paddr_to_vaddr(info, page_middle);
+ if (!readmem(info, page_middle, &pmd_pte, sizeof pmd_pte)) {
+ ERRMSG("Can't get pmd_pte (page_middle:%lx).\n", page_middle);
+ return paddr;
+ }
+
+ /*
+ * Get PTE
+ */
+ temp = vaddr & MASK_PTE;
+ temp = temp >> (PAGE_SHIFT - 3);
+ page_table = pmd_pte + temp;
+ /*
+ * Convert physical address to virtual address
+ */
+ page_table = paddr_to_vaddr(info, page_table);
+ if (!readmem(info, page_table, &pte, sizeof pte)) {
+ ERRMSG("Can't get pte (page_table:%lx).\n", page_table);
+ return paddr;
+ }
+
+ /*
+ * Get physical address
+ */
+ temp = vaddr & MASK_POFFSET;
+ paddr = (pte & _PAGE_PPN_MASK) + temp;
+ if (info->flag_debug) {
+ MSG("vaddr:%llx -> paddr:%lx\n", vaddr, paddr);
+ }
+
+ return paddr;
+}
+
+unsigned long
+ia64_vtop4(struct DumpInfo *info, unsigned long long vaddr)
+{
+ unsigned long paddr = 0, temp, page_dir, pgd_pte, page_upper,
+ pud_pte, page_middle, pmd_pte;
+ unsigned long page_table, pte;
/*
* Translate a virtual address to a physical address
@@ -76,8 +153,8 @@ ia64_vtop(struct DumpInfo *info, unsigne
/*
* Get PGD
*/
- temp = vaddr & MASK_PGD;
- temp = temp >> (PGDIR_SHIFT - 3);
+ temp = vaddr & MASK_PGD_4L;
+ temp = temp >> (PGDIR_SHIFT_4L - 3);
page_dir = SYMBOL(swapper_pg_dir) + temp;
if (!readmem(info, page_dir, &pgd_pte, sizeof pgd_pte)) {
ERRMSG("Can't get pgd_pte (page_dir:%lx).\n", page_dir);
@@ -85,11 +162,26 @@ ia64_vtop(struct DumpInfo *info, unsigne
}
/*
+ * Get PUD
+ */
+ temp = vaddr & MASK_PUD;
+ temp = temp >> (PUD_SHIFT - 3);
+ page_upper = pgd_pte + temp;
+ /*
+ * Convert physical address to virtual address
+ */
+ page_upper = paddr_to_vaddr(info, page_upper);
+ if (!readmem(info, page_upper, &pud_pte, sizeof pud_pte)) {
+ ERRMSG("Can't get pud_pte (page_upper:%lx).\n", page_upper);
+ return paddr;
+ }
+
+ /*
* Get PMD
*/
temp = vaddr & MASK_PMD;
temp = temp >> (PMD_SHIFT - 3);
- page_middle = pgd_pte + temp;
+ page_middle = pud_pte + temp;
/*
* Convert physical address to virtual address
*/
@@ -126,6 +218,30 @@ ia64_vtop(struct DumpInfo *info, unsigne
return paddr;
}
+unsigned long
+ia64_vtop(struct DumpInfo *info, unsigned long long vaddr)
+{
+ unsigned long paddr = 0;
+
+ if (VADDR_REGION(vaddr) != KERNEL_VMALLOC_REGION) {
+ ERRMSG("vaddr(%llx) is not KERNEL_VMALLOC_REGION.\n", vaddr);
+ return paddr;
+ }
+ paddr = vaddr_to_paddr(info, vaddr);
+ if (paddr)
+ return paddr;
+
+ if (!is_vmalloc_addr_ia64(info, vaddr)) {
+ paddr = vaddr - info->kernel_start +
+ (info->phys_base & KERNEL_TR_PAGE_MASK);
+ return paddr;
+ }
+
+ if (info->mem_flags & MEMORY_4LAYER_PAGETABLE)
+ return ia64_vtop4(info, vaddr);
+ else
+ return ia64_vtop3(info, vaddr);
+}
/*
* Convert Virtual Address to File Offest.
@@ -170,5 +286,21 @@ vaddr_to_offset_ia64(struct DumpInfo *in
return offset;
}
+int
+get_machdep_kernel_start_ia64(struct DumpInfo *info)
+{
+ if (SYMBOL(_stext) == NOT_FOUND_SYMBOL)
+ return FALSE;
+
+ info->kernel_start = SYMBOL(_stext);
+
+ if (VADDR_REGION(info->kernel_start) == KERNEL_VMALLOC_REGION)
+ info->vmalloc_start = info->kernel_start + 4*1024UL*1024UL*1024UL;
+ else
+ info->vmalloc_start = KERNEL_VMALLOC_BASE;
+
+ return TRUE;
+}
+
#endif /* ia64 */
--- a/makedumpfile.c
+++ b/makedumpfile.c
@@ -1420,6 +1420,174 @@ get_symbol_info(struct DumpInfo *info)
return TRUE;
}
+
+int
+read_kernel_config(struct DumpInfo *info)
+{
+ int ii, ret, end, found=0;
+ unsigned long size, bufsz;
+ char *pos, *ln, *buf, *head, *tail, *val, *uncomp;
+ char line[512];
+ z_stream stream;
+ unsigned long kernel_config_data;
+
+ kernel_config_data = get_symbol_addr(info, "kernel_config_data");
+ if (kernel_config_data <= 0) {
+ ERRMSG("Can't read kernel cofiguration from kernel binary");
+ return FALSE;
+ }
+
+ /* We don't know how large IKCONFIG is, so we start with
+ * 32k, if we can't find MAGIC_END assume we didn't read
+ * enough, double it and try again.
+ */
+ ii = 32;
+
+again:
+ size = ii * 1024;
+
+ if ((buf = (char *)malloc(size)) == NULL) {
+ MSG("cannot malloc IKCONFIG input buffer\n");
+ return FALSE;
+ }
+
+ if (!readmem(info, kernel_config_data, buf, size)) {
+ MSG("cannot read kernel_config_data\n");
+ goto out2;
+ }
+
+ /* Find the start */
+ if (strstr(buf, MAGIC_START))
+ head = buf + MAGIC_SIZE + 10; /* skip past MAGIC_START and gzip header */
+ else {
+ MSG("could not find MAGIC_START!\n");
+ goto out2;
+ }
+
+ tail = head;
+
+ end = strlen(MAGIC_END);
+
+ /* Find the end*/
+ while (tail < (buf + (size - 1))) {
+
+ if (strncmp(tail, MAGIC_END, end)==0) {
+ found = 1;
+ break;
+ }
+ tail++;
+ }
+
+ if (found) {
+ bufsz = tail - head;
+ size = 10 * bufsz;
+ if ((uncomp = (char *)malloc(size)) == NULL) {
+ MSG("cannot malloc IKCONFIG output buffer\n");
+ goto out2;
+ }
+ } else {
+ if (ii > 512) {
+ MSG("could not find MAGIC_END!\n");
+ goto out2;
+ } else {
+ free(buf);
+ ii *= 2;
+ goto again;
+ }
+ }
+
+
+ /* initialize zlib */
+ stream.next_in = (Bytef *)head;
+ stream.avail_in = (uInt)bufsz;
+
+ stream.next_out = (Bytef *)uncomp;
+ stream.avail_out = (uInt)size;
+
+ stream.zalloc = NULL;
+ stream.zfree = NULL;
+ stream.opaque = NULL;
+
+ ret = inflateInit2(&stream, -MAX_WBITS);
+ if (ret != Z_OK) {
+ ERRMSG("error while reading kernel config, inflateInit2 "
+ "returned %d\n", ret);
+ goto out1;
+ }
+
+ ret = inflate(&stream, Z_FINISH);
+
+ if (ret != Z_STREAM_END) {
+ inflateEnd(&stream);
+ if (ret == Z_NEED_DICT ||
+ (ret == Z_BUF_ERROR && stream.avail_in == 0)) {
+ ERRMSG("error while reading kernel config, stream.avail_in = 0,"
+ "inflate returned %d\n", ret);
+ goto out1;
+ }
+ ERRMSG("error while reading kernel config, inflate returned"
+ "with %d\n", ret);
+ goto out1;
+ }
+ size = stream.total_out;
+
+ ret = inflateEnd(&stream);
+
+ pos = uncomp;
+
+ do {
+ ret = sscanf(pos, "%511[^\n]\n%n", line, &ii);
+ if (ret > 0) {
+ pos += ii;
+
+ ln = line;
+
+ /* skip leading whitespace */
+ while (is_blank(*ln))
+ ln++;
+
+ /* skip comments */
+ if (*ln == '#')
+ continue;
+
+ /* Find '=' */
+ if ((head = strchr(ln, '=')) != NULL) {
+ *head = '\0';
+ val = head + 1;
+
+ head--;
+
+ /* skip trailing whitespace */
+ while (is_blank(*head)) {
+ *head = '\0';
+ head--;
+ }
+
+ /* skip whitespace */
+ while (is_blank(*val))
+ val++;
+
+ } else /* Bad line, skip it */
+ continue;
+
+ if (strcmp(ln, "CONFIG_PGTABLE_4") == 0)
+ info->mem_flags |= MEMORY_4LAYER_PAGETABLE;
+ }
+ } while (ret > 0);
+
+
+ free(uncomp);
+ free(buf);
+ return TRUE;
+
+out1:
+ free(uncomp);
+out2:
+ free(buf);
+
+ return FALSE;
+}
+
int
get_structure_info(struct DumpInfo *info)
{
@@ -2289,9 +2457,15 @@ initial(struct DumpInfo *info)
return FALSE;
}
+ if (!get_machdep_kernel_start(info))
+ return FALSE;
+
if (!check_release(info))
return FALSE;
+ if (!read_kernel_config(info))
+ return FALSE;
+
if (!get_machdep_info(info))
return FALSE;
--- a/makedumpfile.h
+++ b/makedumpfile.h
@@ -29,6 +29,7 @@
#include <libelf.h>
#include <dwarf.h>
#include <byteswap.h>
+#include <ctype.h>
#include "diskdump_mod.h"
/*
@@ -64,6 +65,19 @@ enum {
#define LSEEKED_PDESC (2)
#define LSEEKED_PDATA (3)
+/*
+ * Flags
+ */
+#define MEMORY_4LAYER_PAGETABLE (1 << 0)
+
+/*
+ * For kernel configuration
+ */
+#define MAGIC_START "IKCFG_ST"
+#define MAGIC_END "IKCFG_ED"
+#define MAGIC_SIZE (sizeof(MAGIC_START) - 1)
+
+
static inline int
test_bit(int nr, unsigned long addr)
{
@@ -73,6 +87,12 @@ test_bit(int nr, unsigned long addr)
return ((mask & addr) != 0);
}
+static inline int
+is_blank(int c)
+{
+ return c == ' ' || c == '\t';
+}
+
#define isLRU(flags) test_bit(PG_lru, flags)
#define isPrivate(flags) test_bit(PG_private, flags)
#define isSwapCache(flags) test_bit(PG_swapcache, flags)
@@ -376,12 +396,21 @@ do { \
#define PTRS_PER_PTD_SHIFT (PAGE_SHIFT - 3)
#define PMD_SHIFT (PAGE_SHIFT + PTRS_PER_PTD_SHIFT)
-#define PGDIR_SHIFT (PMD_SHIFT + PTRS_PER_PTD_SHIFT)
+#define PGDIR_SHIFT_3L (PMD_SHIFT + PTRS_PER_PTD_SHIFT)
#define MASK_POFFSET ((1UL << PAGE_SHIFT) - 1)
#define MASK_PTE ((1UL << PMD_SHIFT) - 1) &~((1UL << PAGE_SHIFT) - 1)
-#define MASK_PMD ((1UL << PGDIR_SHIFT) - 1) &~((1UL << PMD_SHIFT) - 1)
-#define MASK_PGD ((1UL << REGION_SHIFT) - 1) & (~((1UL << PGDIR_SHIFT) - 1))
+#define MASK_PMD ((1UL << PGDIR_SHIFT_3L) - 1) &~((1UL << PMD_SHIFT) - 1)
+#define MASK_PGD_3L ((1UL << REGION_SHIFT) - 1) & (~((1UL << PGDIR_SHIFT_3L) - 1))
+
+/*
+ * Layer 4 paging
+ */
+#define PUD_SHIFT (PMD_SHIFT + PTRS_PER_PTD_SHIFT)
+#define PGDIR_SHIFT_4L (PUD_SHIFT + PTRS_PER_PTD_SHIFT)
+
+#define MASK_PUD ((1UL << REGION_SHIFT) - 1) & (~((1UL << PUD_SHIFT) - 1))
+#define MASK_PGD_4L ((1UL << REGION_SHIFT) - 1) & (~((1UL << PGDIR_SHIFT_4L) - 1))
#endif /* ia64 */
@@ -393,6 +422,7 @@ int get_machdep_info_x86();
#define get_phys_base(X) TRUE
#define get_machdep_info(X) get_machdep_info_x86(X)
#define vaddr_to_offset(X, Y) vaddr_to_offset_general(X,Y)
+#define get_machdep_kernel_start(X) TRUE
#endif /* x86 */
#ifdef __x86_64__
@@ -402,6 +432,7 @@ off_t vaddr_to_offset_x86_64();
#define get_phys_base(X) get_phys_base_x86_64(X)
#define get_machdep_info(X) get_machdep_info_x86_64(X)
#define vaddr_to_offset(X, Y) vaddr_to_offset_x86_64(X, Y)
+#define get_machdep_kernel_start(X) TRUE
#endif /* x86_64 */
#ifdef __powerpc__ /* powerpc */
@@ -409,16 +440,19 @@ int get_machdep_info_ppc64();
#define get_machdep_info(X) get_machdep_info_ppc64(X)
#define get_phys_base(X) TRUE
#define vaddr_to_offset(X, Y) vaddr_to_offset_general(X, Y)
+#define get_machdep_kernel_start(X) TRUE
#endif /* powerpc */
#ifdef __ia64__ /* ia64 */
int get_phys_base_ia64();
int get_machdep_info_ia64();
+int get_machdep_kernel_start_ia64();
off_t vaddr_to_offset_ia64();
#define get_machdep_info(X) get_machdep_info_ia64(X)
#define get_phys_base(X) get_phys_base_ia64(X)
#define vaddr_to_offset(X, Y) vaddr_to_offset_ia64(X, Y)
#define VADDR_REGION(X) (((unsigned long)(X)) >> REGION_SHIFT)
+#define get_machdep_kernel_start(X) get_machdep_kernel_start_ia64(X)
#endif /* ia64 */
#define MSG(x...) fprintf(stderr, x)
@@ -509,6 +543,8 @@ struct DumpInfo {
unsigned long max_physmem_bits;
unsigned long sections_per_root;
unsigned long phys_base;
+ unsigned long kernel_start;
+ unsigned long vmalloc_start;
/*
* diskdimp info:
@@ -532,6 +568,7 @@ struct DumpInfo {
*/
unsigned int num_mem_map;
struct mem_map_data *mem_map_data;
+ unsigned int mem_flags;
/*
* Dump memory image info:
