Skip to contentSkip to conversation

IA64: kexec seg fault at xrealloc

Fastboot Akiyama, Nobuyuki 4 participants 5 messages Dec 4, 2006 – Dec 13, 2006

lore
Share & tools
lore
#0 lore raw

Hi Jay,

The attacked patch fixes the problem.
Please try.

kexec/arch/ia64/crashdump-ia64.c +4 -0
diff -Nurp kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c
@@ -316,9 +316,13 @@ int load_crashdump_segments(struct kexec
 	int nr_ranges;
 	size_t size;
 	void *tmp;
+	long int nr_cpus = 0;
+	if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+		return -1;
 	if (info->kexec_flags & KEXEC_ON_CRASH ) {
 		if (get_crash_memory_ranges(&mem_range, &nr_ranges) = 0) {
 			size =  sizeof(Elf64_Ehdr) +
+				(nr_cpus + 1) * sizeof(Elf64_Phdr) +
 				(nr_ranges + 1) * sizeof(Elf64_Phdr);
 			size = (size + EFI_PAGE_SIZE - 1) & ~(EFI_PAGE_SIZE - 1);
 			tmp = xmalloc(size);

Thanks,
Akiyama, Nobuyuki

Quoted 69 lines · On Fri, 01 Dec 2006 15:55:27 -0800 Jay Lan <[email protected]> wrote:

The kexec seg faulted when i ran test at an 56p SN machine. It was successful on a 2p SN.

(gdb) bt

#0  0x200000000016a900 in _int_realloc () from /lib/libc.so.6.1
#1  0x200000000016e020 in realloc () from /lib/libc.so.6.1
#2  0x40000000000020c0 in xrealloc (ptr=0x600000000002ada0, size0)
    at kexec/kexec.c:70
#3  0x40000000000042a0 in add_segment (info=0x60000ffffe2c3718,
    buf=0x600000000002ae30, bufsz288, base 6963621888, memsz384)
    at kexec/kexec.c:310
#4  0x40000000000047f0 in add_buffer (info=0x60000ffffe2c3718,
    buf=0x600000000002ae30, bufsz288, memsz384, buf_align@96,
    buf_min=0, buf_max446744073709551615, buf_end=-1) at
kexec/kexec.c:345
#5  0x400000000001aa20 in load_crashdump_segments (info=0x60000ffffe2c3718,
    ehdr=0x60000ffffe2c3578, max_addr446744073709551615, min_base=0,
    cmdline=0x60000ffffe2c35e8) at kexec/arch/ia64/crashdump-ia64.c:328
#6  0x4000000000016970 in elf_ia64_load (argc=6, argv=0x60000ffffe2c3af8,
    buf=0x2000000000324010 "\177ELF\002\001\001", len939392,
    info=0x60000ffffe2c3718) at kexec/arch/ia64/kexec-elf-ia64.c:203
#7  0x4000000000006a00 in my_load (type=0x0, fileind=5, argc=6,
    argv=0x60000ffffe2c3af8, kexec_flags=1) at kexec/kexec.c:617
#8  0x4000000000008220 in main (argc=6, argv=0x60000ffffe2c3af8)
    at kexec/kexec.c:859

(gdb)

The add_segment() contains code as below:

add_segment()

{
        ...
        last = base + memsz -1;
        if (!valid_memory_range(base, last)) {
                die("Invalid memory segment %p - %p\n",
                        (void *)base, (void *)last);
        }

        size = (info->nr_segments + 1) * sizeof(info->segment[0]);
        info->segment = xrealloc(info->segment, size); <=== seg fault
        info->segment[info->nr_segments].buf   = buf;
        info->segment[info->nr_segments].bufsz = bufsz;
        info->segment[info->nr_segments].mem   = (void *)base;
        info->segment[info->nr_segments].memsz = memsz;

        info->nr_segments++;
        ...
}

The seg fault happened on nr_segment=4. At the end of nr_segment=3 info->segment was set to 2ada0 in the statement

        info->segment = xrealloc(info->segment, size);

And the subsequent call to xrealloc() on nr_segment=4, it died.

On the 2p machine that was successful, the info->segment was set to 29310 at the end of nr_segment=3. It went all the way to 30b10 at the end of 9th segment. So, the value 2ada0 seems still in the bound.

Can anyone more familiar with xrealloc and kexec tell me what might cause the seg fault?

Thanks,
- jay
_______________________________________________
fastboot mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/fastboot

#1 lore raw
Quoted 25 lines · Nobuyuki

Hi Jay,

The attacked patch fixes the problem.
Please try.

diff -Nurp kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c
--- kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c

2006-11-09 19:40:52.000000000 +0900

+++ kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c 2006-11-13 19:17:15.000000000 +0900

kexec/arch/ia64/crashdump-ia64.c +4 -0
@@ -316,9 +316,13 @@ int load_crashdump_segments(struct kexec
        int nr_ranges;
        size_t size;
        void *tmp;
+       long int nr_cpus = 0;
+       if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+               return -1;
        if (info->kexec_flags & KEXEC_ON_CRASH ) {
                if (get_crash_memory_ranges(&mem_range, &nr_ranges) =
 0) {
                        size =  sizeof(Elf64_Ehdr) +
+                               (nr_cpus + 1) * sizeof(Elf64_Phdr) +
                                (nr_ranges + 1) * sizeof(Elf64_Phdr);
                        size = (size + EFI_PAGE_SIZE - 1) &
Quoted 3 lines
~(EFI_PAGE_SIZE - 1);
                        tmp = xmalloc(size);

Hi,
Thanks for finding the is bug, I missed percpu notes segments in size
calculation here. However I think we should also include the nr_ranges in calculation.

	--- a/kexec/arch/ia64/crashdump-ia64.c	2006-12-04 04:19:42.000000000 -0500

+++ b/kexec/arch/ia64/crashdump-ia64.c 2006-12-04 04:21:23.000000000 -0500
@@ -313,13 +313,15 @@ int load_crashdump_segments(struct kexec
{
//struct memory_range *mem_range, *memmap_p;
struct memory_range *mem_range;
- int nr_ranges;
+ int nr_ranges, nr_cpus;
size_t size;
void *tmp;
+ if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+ return -1;
if (info->kexec_flags & KEXEC_ON_CRASH ) {
if (get_crash_memory_ranges(&mem_range, &nr_ranges) = 0) {
size = sizeof(Elf64_Ehdr) +
- (nr_ranges + 1) * sizeof(Elf64_Phdr);
+ (nr_ranges + nr_cpus + 1) * sizeof(Elf64_Phdr);
size = (size + EFI_PAGE_SIZE - 1) & ~(EFI_PAGE_SIZE - 1); tmp = xmalloc(size);
memset(tmp, 0, size);

Thanks
Zou Nan hai

Quoted 3 lines

Thanks,
Akiyama, Nobuyuki

#2 lore raw
In reply to Akiyama, Nobuyuki

Hi Nan hai,

Thanks for clean-up.
Your code is better than mine;-)

Thanks,
Akiyama, Nobuyuki

Quoted 64 lines · On 04 Dec 2006 14:48:33 +0800 Zou Nan hai <[email protected]> wrote:
Quoted 25 lines · Nobuyuki

Hi Jay,

The attacked patch fixes the problem.
Please try.

diff -Nurp kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c
--- kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c

2006-11-09 19:40:52.000000000 +0900

+++ kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c 2006-11-13 19:17:15.000000000 +0900

kexec/arch/ia64/crashdump-ia64.c +4 -0
@@ -316,9 +316,13 @@ int load_crashdump_segments(struct kexec
        int nr_ranges;
        size_t size;
        void *tmp;
+       long int nr_cpus = 0;
+       if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+               return -1;
        if (info->kexec_flags & KEXEC_ON_CRASH ) {
                if (get_crash_memory_ranges(&mem_range, &nr_ranges) =
 0) {
                        size =  sizeof(Elf64_Ehdr) +
+                               (nr_cpus + 1) * sizeof(Elf64_Phdr) +
                                (nr_ranges + 1) * sizeof(Elf64_Phdr);
                        size = (size + EFI_PAGE_SIZE - 1) &
Quoted 3 lines
~(EFI_PAGE_SIZE - 1);
                        tmp = xmalloc(size);

Hi,
Thanks for finding the is bug, I missed percpu notes segments in size
calculation here. However I think we should also include the nr_ranges in calculation.

	--- a/kexec/arch/ia64/crashdump-ia64.c	2006-12-04 04:19:42.000000000 -0500

+++ b/kexec/arch/ia64/crashdump-ia64.c 2006-12-04 04:21:23.000000000 -0500

kexec/arch/ia64/crashdump-ia64.c +4 -2
@@ -313,13 +313,15 @@ int load_crashdump_segments(struct kexec
 {
 	//struct memory_range *mem_range, *memmap_p;
 	struct memory_range *mem_range;
-	int nr_ranges;
+	int nr_ranges, nr_cpus;
 	size_t size;
 	void *tmp;
+        if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+		return -1;
 	if (info->kexec_flags & KEXEC_ON_CRASH ) {
 		if (get_crash_memory_ranges(&mem_range, &nr_ranges) = 0) {
 			size =  sizeof(Elf64_Ehdr) +
-				(nr_ranges + 1) * sizeof(Elf64_Phdr);
+				(nr_ranges + nr_cpus + 1) * sizeof(Elf64_Phdr);
 			size = (size + EFI_PAGE_SIZE - 1) & ~(EFI_PAGE_SIZE - 1);
 			tmp = xmalloc(size);
 			memset(tmp, 0, size);

Thanks
Zou Nan hai

Quoted 3 lines

Thanks,
Akiyama, Nobuyuki

#3 lore raw
In reply to Akiyama, Nobuyuki

Hi Nan-hai and Akiyama San,

Thanks for your patch. Nan-hai's patch fixed the problem!

Cheers,
- jay

Quoted 68 lines · Zou Nan hai
Quoted 25 lines · Nobuyuki

Hi Jay,

The attacked patch fixes the problem.
Please try.

diff -Nurp kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c
--- kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c

2006-11-09 19:40:52.000000000 +0900

+++ kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c 2006-11-13 19:17:15.000000000 +0900

kexec/arch/ia64/crashdump-ia64.c +4 -0
@@ -316,9 +316,13 @@ int load_crashdump_segments(struct kexec
        int nr_ranges;
        size_t size;
        void *tmp;
+       long int nr_cpus = 0;
+       if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+               return -1;
        if (info->kexec_flags & KEXEC_ON_CRASH ) {
                if (get_crash_memory_ranges(&mem_range, &nr_ranges) =
 0) {
                        size =  sizeof(Elf64_Ehdr) +
+                               (nr_cpus + 1) * sizeof(Elf64_Phdr) +
                                (nr_ranges + 1) * sizeof(Elf64_Phdr);
                        size = (size + EFI_PAGE_SIZE - 1) &
Quoted 3 lines
~(EFI_PAGE_SIZE - 1);
                        tmp = xmalloc(size);

Hi,
Thanks for finding the is bug, I missed percpu notes segments in size
calculation here. However I think we should also include the nr_ranges in calculation.

	--- a/kexec/arch/ia64/crashdump-ia64.c	2006-12-04 04:19:42.000000000 -0500

+++ b/kexec/arch/ia64/crashdump-ia64.c 2006-12-04 04:21:23.000000000 -0500

kexec/arch/ia64/crashdump-ia64.c +4 -2
@@ -313,13 +313,15 @@ int load_crashdump_segments(struct kexec
 {
 	//struct memory_range *mem_range, *memmap_p;
 	struct memory_range *mem_range;
-	int nr_ranges;
+	int nr_ranges, nr_cpus;
 	size_t size;
 	void *tmp;
+        if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+		return -1;
 	if (info->kexec_flags & KEXEC_ON_CRASH ) {
 		if (get_crash_memory_ranges(&mem_range, &nr_ranges) = 0) {
 			size =  sizeof(Elf64_Ehdr) +
-				(nr_ranges + 1) * sizeof(Elf64_Phdr);
+				(nr_ranges + nr_cpus + 1) * sizeof(Elf64_Phdr);
 			size = (size + EFI_PAGE_SIZE - 1) & ~(EFI_PAGE_SIZE - 1);
 			tmp = xmalloc(size);
 			memset(tmp, 0, size);

Thanks
Zou Nan hai

Quoted 3 lines

Thanks,
Akiyama, Nobuyuki

-
To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#4 lore raw
In reply to Akiyama, Nobuyuki
Quoted 45 lines · Zou Nan hai
Quoted 25 lines · Nobuyuki

Hi Jay,

The attacked patch fixes the problem.
Please try.

diff -Nurp kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c
--- kexec-tools-1.101.org/kexec/arch/ia64/crashdump-ia64.c

2006-11-09 19:40:52.000000000 +0900

+++ kexec-tools-1.101/kexec/arch/ia64/crashdump-ia64.c 2006-11-13 19:17:15.000000000 +0900

kexec/arch/ia64/crashdump-ia64.c +4 -0
@@ -316,9 +316,13 @@ int load_crashdump_segments(struct kexec
        int nr_ranges;
        size_t size;
        void *tmp;
+       long int nr_cpus = 0;
+       if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+               return -1;
        if (info->kexec_flags & KEXEC_ON_CRASH ) {
                if (get_crash_memory_ranges(&mem_range, &nr_ranges) =
 0) {
                        size =  sizeof(Elf64_Ehdr) +
+                               (nr_cpus + 1) * sizeof(Elf64_Phdr) +
                                (nr_ranges + 1) * sizeof(Elf64_Phdr);
                        size = (size + EFI_PAGE_SIZE - 1) &
Quoted 3 lines
~(EFI_PAGE_SIZE - 1);
                        tmp = xmalloc(size);

Hi,
Thanks for finding the is bug, I missed percpu notes segments in size
calculation here. However I think we should also include the nr_ranges in calculation.

	--- a/kexec/arch/ia64/crashdump-ia64.c	2006-12-04 04:19:42.000000000 -0500

+++ b/kexec/arch/ia64/crashdump-ia64.c 2006-12-04 04:21:23.000000000 -0500

kexec/arch/ia64/crashdump-ia64.c +1 -1
@@ -313,13 +313,15 @@ int load_crashdump_segments(struct kexec
 {
 	//struct memory_range *mem_range, *memmap_p;
 	struct memory_range *mem_range;
-	int nr_ranges;
+	int nr_ranges, nr_cpus;
 	size_t size;
 	void *tmp;

kexec/arch/ia64/crashdump-ia64.c int load_crashdump_segments(struct kexec · line 327

Quoted 10 lines
kexec/arch/ia64/crashdump-ia64.c +3 -1
+        if ((nr_cpus = sysconf(_SC_NPROCESSORS_CONF)) < 0)
+		return -1;
 	if (info->kexec_flags & KEXEC_ON_CRASH ) {
 		if (get_crash_memory_ranges(&mem_range, &nr_ranges) = 0) {
 			size =  sizeof(Elf64_Ehdr) +
-				(nr_ranges + 1) * sizeof(Elf64_Phdr);
+				(nr_ranges + nr_cpus + 1) * sizeof(Elf64_Phdr);
 			size = (size + EFI_PAGE_SIZE - 1) & ~(EFI_PAGE_SIZE - 1);
 			tmp = xmalloc(size);
 			memset(tmp, 0, size);

Hi,

that patch looks correct to me. However, I believe that the problem is already resolved in kexec-tools-testing by using the generic /proc/iomem handling code that was introduced in changesets c80198e78ce26783e092645b9ac8587e1374f22f and f038d30bff8510a1df1e72af08db1766581d1f2c.

Could someone please test this?

Signature · 3 lines
Horms
  H: http://www.vergenet.net/~horms/
  W: http://www.valinux.co.jp/en/