m68k 54418 fails to execute user space
Greg Ungerer
gerg at linux-m68k.org
Sun Jun 30 15:35:16 PDT 2024
Hi Michael,
On 28/6/24 17:48, Michael Schmitz wrote:
> Jean-Michel,
>
> Am 28.06.2024 um 19:24 schrieb Jean-Michel Hautbois:
>>> I forgot to take into account that libraries are loaded only the
>>> binary starts executing. Can you print the same map dump in the exit
>>> syscall code path? That ought to show all loaded libraries at that point.
>>
>> Thanks for your suggestion !
>> I changed it a bit, and I added a call in do_exit() as suggested. When
>> executing ls I get:
>>
>> bash-5.2# ls -l
>> load_elf_binary: Dump memory for ls (31):
>> mmap: 60000000-6001e000 r-xp 00000000 00:00 178 /lib/ld.so.1
>> mmap: 6001e000-60022000 rw-p 0001c000 00:00 178 /lib/ld.so.1
>> mmap: 70000000-700c2000 r-xp 00000000 00:00 28 /bin/busybox
>> mmap: 700c2000-700ca000 rw-p 000c0000 00:00 28 /bin/busybox
>> mmap: bfc1e000-bfc40000 rw-p bffde000 00:00 0 [stack]
>>
>> do_exit: Dump memory for ls (31):
>> mmap: 60000000-6001e000 r-xp 00000000 00:00 178 /lib/ld.so.1
>> mmap: 6001e000-60020000 r--p 0001c000 00:00 178 /lib/ld.so.1
>> mmap: 60020000-60022000 rw-p 0001e000 00:00 178 /lib/ld.so.1
>> mmap: 60022000-6002c000 r-xp 00000000 00:00 193 /lib/libresolv.so.2
>> mmap: 6002c000-6002e000 r--p 00008000 00:00 193 /lib/libresolv.so.2
>> mmap: 6002e000-60030000 rw-p 0000a000 00:00 193 /lib/libresolv.so.2
>> mmap: 60030000-60032000 rw-p 60030000 00:00 0
>> mmap: 60032000-6015a000 r-xp 00000000 00:00 185 /lib/libc.so.6
>> mmap: 6015a000-6015c000 r--p 00126000 00:00 185 /lib/libc.so.6
>> mmap: 6015c000-60160000 rw-p 00128000 00:00 185 /lib/libc.so.6
>> mmap: 60160000-6016e000 rw-p 60160000 00:00 0
>> mmap: 70000000-700c2000 r-xp 00000000 00:00 28 /bin/busybox
>> mmap: 700c2000-700c4000 r--p 000c0000 00:00 28 /bin/busybox
>> mmap: 700c4000-700ca000 rw-p 000c2000 00:00 28 /bin/busybox
>> mmap: 700ca000-700ec000 rwxp 700ca000 00:00 0 [heap]
>> mmap: bfc1e000-bfc40000 rw-p bffde000 00:00 0 [stack]
>>
>> When I call it a second time, I get:
>>
>> bash-5.2# ls -l
>> load_elf_binary: Dump memory for ls (33):
>> mmap: 60000000-6001e000 r-xp 00000000 00:00 178 /lib/ld.so.1
>> mmap: 6001e000-60022000 rw-p 0001c000 00:00 178 /lib/ld.so.1
>> mmap: 70000000-700c2000 r-xp 00000000 00:00 28 /bin/busybox
>> mmap: 700c2000-700ca000 rw-p 000c0000 00:00 28 /bin/busybox
>> mmap: bfb5a000-bfb7c000 rw-p bffde000 00:00 0 [stack]
>> do_exit: Dump memory for ls (33):
>> mmap: 60000000-6001e000 r-xp 00000000 00:00 178 /lib/ld.so.1
>> mmap: 6001e000-60020000 r--p 0001c000 00:00 178 /lib/ld.so.1
>> mmap: 60020000-60022000 rw-p 0001e000 00:00 178 /lib/ld.so.1
>> mmap: 60022000-6002c000 r-xp 00000000 00:00 193 /lib/libresolv.so.2
>> mmap: 6002c000-6002e000 r--p 00008000 00:00 193 /lib/libresolv.so.2
>> mmap: 6002e000-60030000 rw-p 0000a000 00:00 193 /lib/libresolv.so.2
>> mmap: 60030000-60032000 rw-p 60030000 00:00 0
>> mmap: 60032000-6015a000 r-xp 00000000 00:00 185 /lib/libc.so.6
>> mmap: 6015a000-6015c000 r--p 00126000 00:00 185 /lib/libc.so.6
>> mmap: 6015c000-60160000 rw-p 00128000 00:00 185 /lib/libc.so.6
>> mmap: 60160000-6016e000 rw-p 60160000 00:00 0
>> mmap: 70000000-700c2000 r-xp 00000000 00:00 28 /bin/busybox
>> mmap: 700c2000-700c4000 r--p 000c0000 00:00 28 /bin/busybox
>> mmap: 700c4000-700ca000 rw-p 000c2000 00:00 28 /bin/busybox
>
> No heap in this second call. Can you print mm->start_brk and mm->brk please?
>
> The process memory layout is a little unusual (I would have expected the binary to be mapped before the dynamic libraries, not after). Is that expected on Coldfire, Greg?
I am not entirely sure of the history behind the layouts. But for the M547x family
I have done most MMU work on this is typical. So like this:
# cat /proc/1/maps
60000000-60008000 r-xp 00000000 1f:00 550544 /lib/ld-uClibc-0.9.33.2.so
60008000-6000a000 r--p 00006000 1f:00 550544 /lib/ld-uClibc-0.9.33.2.so
6000a000-6000c000 rw-p 00008000 1f:00 550544 /lib/ld-uClibc-0.9.33.2.so
6000c000-6000e000 rw-p 00000000 00:00 0
60010000-60014000 r-xp 00000000 1f:00 1194384 /lib/libcrypt-0.9.33.2.so
60014000-60016000 r--p 00002000 1f:00 1194384 /lib/libcrypt-0.9.33.2.so
60016000-60018000 rw-p 00004000 1f:00 1194384 /lib/libcrypt-0.9.33.2.so
60018000-60028000 rw-p 00000000 00:00 0
60028000-60080000 r-xp 00000000 1f:00 184160 /lib/libuClibc-0.9.33.2.so
60080000-60082000 r--p 00056000 1f:00 184160 /lib/libuClibc-0.9.33.2.so
60082000-60084000 rw-p 00058000 1f:00 184160 /lib/libuClibc-0.9.33.2.so
60084000-60086000 rw-p 00000000 00:00 0
80000000-80004000 r-xp 00000000 1f:00 1882624 /bin/init
80004000-80006000 r--p 00002000 1f:00 1882624 /bin/init
80006000-80008000 rw-p 00004000 1f:00 1882624 /bin/init
80008000-8000a000 rwxp 00000000 00:00 0 [heap]
bfdba000-bfddc000 rwxp 00000000 00:00 0 [stack]
The 0x60000000 library addresses are due to mmaping - and that is based on the
definition of TASK_UNMAPPED_BASE for ColdFire, from arch/m68k/include/asm/processor.h
/* This decides where the kernel will search for a free chunk of vm
* space during mmap's.
*/
#ifdef CONFIG_MMU
#if defined(CONFIG_COLDFIRE)
#define TASK_UNMAPPED_BASE 0x60000000UL
The application address range of 0x80000000 are baked in at link time:
$ m68k-linux-objdump --headers bin/init
init: file format elf32-m68k
Sections:
Idx Name Size VMA LMA File off Algn
0 .interp 0000000d 80000114 80000114 00000114 2**0
CONTENTS, ALLOC, LOAD, READONLY, DATA
1 .hash 00000170 80000124 80000124 00000124 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
2 .gnu.hash 000001b4 80000294 80000294 00000294 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
3 .dynsym 00000350 80000448 80000448 00000448 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
4 .dynstr 00000174 80000798 80000798 00000798 2**0
CONTENTS, ALLOC, LOAD, READONLY, DATA
5 .rela.dyn 00000018 8000090c 8000090c 0000090c 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
6 .rela.plt 00000240 80000924 80000924 00000924 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
7 .init 00000014 80000b64 80000b64 00000b64 2**1
CONTENTS, ALLOC, LOAD, READONLY, CODE
8 .plt 00000498 80000b78 80000b78 00000b78 2**2
CONTENTS, ALLOC, LOAD, READONLY, CODE
9 .text 00001258 80001010 80001010 00001010 2**2
CONTENTS, ALLOC, LOAD, READONLY, CODE
10 .fini 0000000e 80002268 80002268 00002268 2**1
CONTENTS, ALLOC, LOAD, READONLY, CODE
11 .rodata 00000257 80002276 80002276 00002276 2**0
CONTENTS, ALLOC, LOAD, READONLY, DATA
12 .eh_frame 00000004 800024d0 800024d0 000024d0 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
13 .ctors 00000008 80005f30 80005f30 00003f30 2**2
CONTENTS, ALLOC, LOAD, DATA
14 .dtors 00000008 80005f38 80005f38 00003f38 2**2
CONTENTS, ALLOC, LOAD, DATA
15 .dynamic 000000c0 80005f40 80005f40 00003f40 2**2
CONTENTS, ALLOC, LOAD, DATA
16 .got 000000cc 80006000 80006000 00004000 2**2
CONTENTS, ALLOC, LOAD, DATA
17 .data 00000008 800060cc 800060cc 000040cc 2**2
CONTENTS, ALLOC, LOAD, DATA
18 .bss 0000002c 800060d4 800060d4 000040d4 2**2
ALLOC
I am not sure why JM's link has applications linked at 0x70000000.
Is that a glibc thing? My examples above are all based on uClibc.
While we are talking about link addresses, JM, can you tell me what
your kernel is linked at? For me it is from a base near 0 (well actually
128k offset, but there is some meaning to that address) which matches
the physical DRAM which starts at address 0:
$ m68k-linux-objdump --headers linux/vmlinux
linux/vmlinux: file format elf32-m68k
Sections:
Idx Name Size VMA LMA File off Algn
0 .text 002e1800 00020000 00020000 00002000 2**2
CONTENTS, ALLOC, LOAD, READONLY, CODE
1 .rodata 0003f398 00302000 00302000 002e4000 2**4
CONTENTS, ALLOC, LOAD, DATA
...
I know the 54418 typically has DRAM at a different physical offset
(I think it is 0x40000000?), so wondering if the VM layout was
adjusted in some way to cater for that difference?
Regards
Greg
> Cheers,
>
> Michael
>
>
>> mmap: bfb5a000-bfb7c000 rw-p bffde000 00:00 0 [stack]
>>
>> The first call generates the "ls" output, not the second one.
>> The helper looks like:
>> diff --git a/mm/mmap.c b/mm/mmap.c
>> index 83b4682ec85c..14d861e9cba2 100644
>> --- a/mm/mmap.c
>> +++ b/mm/mmap.c
>> @@ -76,6 +76,87 @@ int mmap_rnd_compat_bits __read_mostly =
>> CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
>> static bool ignore_rlimit_data;
>> core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
>>
>> +int dump_memory_map(struct task_struct *task)
>> +{
>> + struct mm_struct *mm = task->mm;
>> + struct vm_area_struct *vma;
>> + struct file *file;
>> + struct path *path;
>> + char *buf;
>> + char *pathname;
>> +
>> + if (!mm) {
>> + return -ENOMEM;
>> + }
>> +
>> + MA_STATE(mas, &mm->mm_mt, 0, -1);
>> + // Acquire the read lock for mmap_lock
>> + down_read(&mm->mmap_lock);
>> + mas_lock(&mas);
>> + for (vma = mas_find(&mas, ULONG_MAX); vma; vma = mas_find(&mas,
>> ULONG_MAX)) {
>> + char perms[5] = "---p"; // Default permissions
>> + // Set permissions based on vm_flags
>> + if (vma->vm_flags & VM_READ) perms[0] = 'r';
>> + if (vma->vm_flags & VM_WRITE) perms[1] = 'w';
>> + if (vma->vm_flags & VM_EXEC) perms[2] = 'x';
>> + if (vma->vm_flags & VM_MAYSHARE) perms[3] = 's';
>> +
>> + if (vma->vm_file) { // If there's an associated file
>> + buf = (char *)__get_free_page(GFP_KERNEL);
>> + if (!buf) {
>> + continue; // Handle memory allocation
>> failure
>> + }
>> +
>> + file = vma->vm_file;
>> + path = &file->f_path;
>> + pathname = d_path(path, buf, PAGE_SIZE);
>> + if (IS_ERR(pathname)) {
>> + pathname = NULL;
>> + }
>> +
>> + // Print memory area information with file path
>> + pr_info("%08lx-%08lx %s %08lx %02x:%02x %lu %s\n",
>> + vma->vm_start, vma->vm_end,
>> + perms,
>> + vma->vm_pgoff << PAGE_SHIFT,
>> + MAJOR(file_inode(file)->i_rdev),
>> + MINOR(file_inode(file)->i_rdev),
>> + file_inode(file)->i_ino,
>> + pathname ? pathname : "");
>> +
>> + free_page((unsigned long)buf);
>> + } else {
>> + char *special_area_name = NULL;
>> +
>> + // Check for heap
>> + if (vma->vm_end > mm->start_brk && vma->vm_start
>> < mm->brk) {
>> + special_area_name = "[heap]";
>> + }
>> + // Check for stack
>> + else if (vma->vm_start <= mm->start_stack &&
>> vma->vm_end >= mm->start_stack) {
>> + special_area_name = "[stack]";
>> + }
>> + // Check for vdso
>> + else if (vma->vm_flags & VM_EXEC &&
>> vma->vm_flags & VM_READ && !vma->vm_file) {
>> + special_area_name = "[vdso]";
>> + }
>> +
>> + // Print memory area information without file path
>> + pr_info("%08lx-%08lx %s %08lx 00:00 0 %s\n",
>> + vma->vm_start, vma->vm_end,
>> + perms,
>> + vma->vm_pgoff << PAGE_SHIFT,
>> + special_area_name ? special_area_name :
>> " ");
>> + }
>> + }
>> + mas_unlock(&mas);
>> + // Release the read lock for mmap_lock
>> + up_read(&mm->mmap_lock);
>> +
>> + return 0;
>> +}
>> +EXPORT_SYMBOL(dump_memory_map);
>>
>>
>> Thanks,
>> JM
More information about the linux-mtd
mailing list