[lkml]   [2011]   [Mar]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: "x86-64, mm: Put early page table high" causes crash on Xen
    On Wed, 2 Mar 2011, Konrad Rzeszutek Wilk wrote:
    > On Wed, Mar 02, 2011 at 03:23:48PM +0000, Stefano Stabellini wrote:
    > > On Tue, 1 Mar 2011, Yinghai Lu wrote:
    > > > We do need map_low_page ( BTW, that name is totally misleading...)
    > > >
    > > > the reason is we put page_table high and at that time is not under max_pfn_mapped. (aka not mapped).
    > > >
    > > > So have to use
    > > > adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE);
    > > > to early map it and Read/Write to it.
    > >
    > > I think I have figured out a way to update the logic of mask_rw_pte to
    > > account for the new way of allocating kernel pagetables.
    > > The appended patch fix the boot crash for me.
    > >
    > > ---
    > >
    > > xen: update mask_rw_pte after kernel page tables init changes
    > >
    > > Already existing kernel page table pages can now be mapped using
    > > early_ioremap too so we need to update mask_rw_pte to make sure these
    > > pages are still mapped RO.
    > > We do that by mapping RO all the pages mapped using early_ioremap apart
    > > from the last one that has been allocated because it is not a page table
    > > page yet (it has been hooked into the page tables yet).
    > ^- has not?
    > ..this is b/c the initial_kernel_mapping and its family of calls update
    > the pgt_buf_end (used to be called e820_table_end), _before__
    > it calls the Xen MMU to set the PTE entries and then after it is done
    > update the PMD?. Hence the simple check to see if the PFN is the _old_
    > value of the pgt_buf_end and if so skip altering the mapping from RW to RO
    > and leave them be?

    yes and yes

    > On subsequent passes we fall within the first conditional
    > and update the PTE to RO?
    > When is that pass done?

    kernel_physical_mapping_init doesn't go through the same pagetable
    pages twice so that is not the issue.
    The problem is that kernel_physical_mapping_init is now re-mapping all
    the initial pagetable pages using early_ioremap, so the usage of
    early_ioremap is not enough anymore to distinguish between freshly
    allocated page table pages and already hooked pagetable pages.
    I am investigating the possibility that the introduction of map_low_page
    is actually overkill for this reason.

    > You also might want to mention the git commits that inspired this patch
    > and include the nice description you provided in the first email of this
    > thread.


     \ /
      Last update: 2011-03-07 17:11    [W:0.021 / U:50.628 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site