On 12/28/15 at 02:32pm, Xunlei Pang wrote:
> On 12/24/2015 at 02:44 PM, Xunlei Pang wrote:
> >>>>> +static void kexec_mark_crashkres(bool protect)
> >>>>> +{
> >>>>> + unsigned long control;
> >>>>> +
> >>>>> + kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
> >>>>> +
> >>>>> + /* Don't touch the control code page used in crash_kexec().*/
> >>>>> + control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
> >>>>> + /* Control code page is located in the 2nd page. */
> >>>>> + control = control + PAGE_SIZE;
> >> Though it works because the control code is less than 1 page, but use the macro
> >> of KEXEC_CONTROL_PAGE_SIZE looks better..
>
> The 1st page is pagetable, control code page locates at the 2nd page.
> The following kexec_mark_range() wants to mark ro from crashk_res.start
> to the 1st page(included), so here we must use PAGE_SIZE.
>
> >>
> >>>>> + kexec_mark_range(crashk_res.start, control - 1, protect);
> >>>>> + kexec_mark_range(control + PAGE_SIZE, crashk_res.end, protect);
> >>>> X86 kexec will copy the page while kexecing, could you check if we can move
> >>>> that copying to earliyer while kexec loading, maybe machine_kexec_prepare so
> >>>> that we can make a arch-independent implementation.
> >>> For some arch, may use huge tlb directly to do the kernel mapping,
> >>> in such cases, we can't implement this function. So I think it should
> >>> be arch-dependent.
> >> Ok, that's fine.
> > At least moving the x86 control-copying code into arch-related
> > machine_kexec_prepare() should work, and this can omit the
> > special treatment of the control code page.
>
> The "relocate_kernel" routine in "relocate_kernel_64.S" will use it as
> a temp storage "for jumping back"(as its comment), so we can't mark
> it readonly.
kexec will copy the relocate_kernel binary to control_code_page in
function machine_kexec. This is a major reason to set the region
control_code_page to control_code_page + PAGE_SIZE with mode read/write.
Thanks
Minfei
