Re: [PATCH v2] x86_64, vmcoreinfo: Append 'page_offset_base' to vmcoreinfo

From: Baoquan He
Date: Tue Nov 27 2018 - 18:29:23 EST


On 11/27/18 at 02:16pm, Kees Cook wrote:
> Why is KERNELOFFSET= not sufficient?
>
> See commit b6085a865762 ("x86, kaslr: export offset in VMCOREINFO ELF notes")
>
> + vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
> + (unsigned long)&_text - __START_KERNEL);

KERNELOFFSET is virtual address delta after kernel text KASLR, namely
the offset from the original default kernel text virtual address,
0xffffffff88000000.

While after memory region KASLR in kernel_randomize_memory(), the
starting address of the direct mapping of physical memory, PAGE_OFFSET,
is changed too. We need get it to analyze memory in makedumpfile/crash.
Currently we deduce it from elf program segment of kcore:
Program Headers:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
......

LOAD 0x00000a62c0004000 0xffff8a62c0001000 0x0000000000001000
0x000000000009c000 0x000000000009c000 RWE 1000

page_offset = 0xffff8a62c0001000 - 0x0000000000001000;
Since we put the direct mapping segments at the bottom part of kcore, we
can always get page_offset right.

Thanks
Baoquan

>
> -Kees
>
> >> diff --git a/arch/x86/kernel/machine_kexec_64.c b/arch/x86/kernel/machine_kexec_64.c
> >> index 4c8acdfdc5a7..6161d77c5bfb 100644
> >> --- a/arch/x86/kernel/machine_kexec_64.c
> >> +++ b/arch/x86/kernel/machine_kexec_64.c
> >> @@ -356,6 +356,9 @@ void arch_crash_save_vmcoreinfo(void)
> >> VMCOREINFO_SYMBOL(init_top_pgt);
> >> vmcoreinfo_append_str("NUMBER(pgtable_l5_enabled)=%d\n",
> >> pgtable_l5_enabled());
> >> +#ifdef CONFIG_RANDOMIZE_BASE
> >> + VMCOREINFO_NUMBER(page_offset_base);
> >> +#endif
> >>
> >> #ifdef CONFIG_NUMA
> >> VMCOREINFO_SYMBOL(node_data);
>
> --
> Kees Cook
>
> _______________________________________________
> kexec mailing list
> kexec@xxxxxxxxxxxxxxxxxxx
> http://lists.infradead.org/mailman/listinfo/kexec