|
Description
|
The x86 CR registers CR0, CR2, CR3 and CR4 should be saved at the time of a fault
for 32-bit systems.
In addition, CR8 should be saved on a core dump on amd64 systems.
These registers should be output to user and kernel core files, and should be
visible in mdb.
Common example:
A user program dies due to a page fault, generates a SIGSEGV, and leaves a core file.
If the faulting instruction was a memory reference off a segment register, there is
no way to determine the faulting address because CR2 is not saved to the core file,
the faulting instruction may show only a reference off a segment register, and only
segment register selectors are saved.
Debugging such a reference as part of a panic is a little easier, but the additional
processor state in the other CR registers may still be helpful.
The information about the actual fault is stored in the pr_info.si_addr field of the lwpstatus info, stored in the core file:
% elfdump -n core
...
entry [11]
namesz: 0x5
descsz: 0x320
type: [ NT_LWPSTATUS ]
name:
CORE\0
desc: (lwpstatus_t)
pr_flags: [ PR_PCINVAL PR_MSACCT PR_MSFORK ]
pr_lwpid: 1
pr_why: 0x0
pr_what: 0x0
pr_cursig: [ SIGSEGV ]
pr_info:
si_signo: [ SIGSEGV ]
si_errno: 0
si_code: [ SEGV_MAPERR ]
si_addr: 0xefd61af0
pr_lwppend: 0
...
(see the si_addr field)
So there is no need to add anything to the core dump to get this information. It would be reasonable to request that mdb be
able to print this information.
|
