Re: "x86-64, mm: Put early page table high" causes crash on Xen

[Yinghai Lu]

On Mon, Mar 7, 2011 at 7:47 AM, Stefano Stabellini
<stefano.stabellini@xxxxxxxxxxxxx> wrote:
> On Tue, 1 Mar 2011, Yinghai Lu wrote:
>> On 03/01/2011 09:21 AM, Stefano Stabellini wrote:
>> > Yinghai Lu,
>> > while testing tip/x86/mm on Xen I found out that the commit "x86-64, mm:
>> > Put early page table high" reliably crashes Linux at boot.
>> > The reason is well explained by the commit message of
>> > fef5ba797991f9335bcfc295942b684f9bf613a1:
>> >
>> > "Xen requires that all pages containing pagetable entries to be mapped
>> > read-only.  If pages used for the initial pagetable are already mapped
>> > then we can change the mapping to RO.  However, if they are initially
>> > unmapped, we need to make sure that when they are later mapped, they
>> > are also mapped RO.
>> >
>> > We do this by knowing that the kernel pagetable memory is pre-allocated
>> > in the range e820_table_start - e820_table_end, so any pfn within this
>> > range should be mapped read-only.  However, the pagetable setup code
>> > (Those mappings are removed before the pages are presented to Xen
>> > as pagetable pages.)"
>> >
>> > In other words mask_rw_pte (called by xen_set_pte_init) should mark RO
>> > the already existing pagetable pages (like the ones belonging to the
>> > initial mappings), while it should mark RW the new pages not yet hooked
>> > into the pagetable.  This is what the following lines used to achieve,
>> > but don't anymore:
>> >
>> >     /*
>> >      * If the new pfn is within the range of the newly allocated
>> >      * kernel pagetable, and it isn't being mapped into an
>> >      */
>> >         pfn >= pgt_buf_start && pfn < pgt_buf_end)
>> >             pte = pte_wrprotect(pte);
>> >
>> > Unfortunately now we map the already existing initial pagetable pages a
>> > second time and the new zeroed pages using map_low_page, so we are
>> > unable to distinguish between the two.
>> >
>> > Can we go back to the previous way of accessing pagetable pages from
>> > kernel_physical_mapping_init, while keeping the new pagetable allocation
>> > strategy? It seems to me that the introduction of map_low_page is not
>> > actually required, is it? In that case we could just revert that bit...
>> > (appended partial revert example).
>>
>> 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.
>
> As you might already know from my other email few days ago, I have found
> a solution to this problem modifying only Xen specific code.
> However I went back to this because I want to be sure that the patch is
> the correct solution rather than a workaround.
>
> It seems to me that only pagetable pages freshly allocated with
> alloc_low_page needs to be mapped because initial pagetable pages are
> under max_pfn_mapped.
> Considering that kernel_physical_mapping_init doesn't go through the
> same pagetable page twice, it is never the case that we map_low_page
> anything but initial pagetable pages that are under max_pfn_mapped
> anyway.
> For example this is what happens on my machine:
>
>
> [    0.000000] initial memory mapped : 0 - 024be000
> [    0.000000] init_memory_mapping: 0000000000000000-00000000cffc2000
> [    0.000000]  0000000000 - 00cffc2000 page 4k
> [    0.000000] kernel direct mapping tables up to cffc2000 @ cf93d000-cffc2000
> [    0.000000] DEBUG map low page(2004000, 00001000)
> [    0.000000] DEBUG map low page(2008000, 00001000)
> [    0.000000] DEBUG map low page(2c62000, 00001000)
> [    0.000000] DEBUG map low page(2c65000, 00001000)
> [    0.000000] DEBUG map low page(2c66000, 00001000)
> [    0.000000] DEBUG map low page(2c67000, 00001000)
> [    0.000000] DEBUG map low page(2c68000, 00001000)
> [    0.000000] DEBUG map low page(2c69000, 00001000)
> [    0.000000] DEBUG map low page(2c6a000, 00001000)
> [    0.000000] DEBUG map low page(2c6b000, 00001000)
> [    0.000000] DEBUG map low page(2c6c000, 00001000)
> [    0.000000] DEBUG map low page(2c6d000, 00001000)
> [    0.000000] DEBUG map low page(2c6e000, 00001000)
> [    0.000000] DEBUG map low page(2c6f000, 00001000)
> [    0.000000] DEBUG map low page(2c70000, 00001000)
> [    0.000000] DEBUG map low page(2c71000, 00001000)
> [    0.000000] DEBUG map low page(2c72000, 00001000)
> [    0.000000] DEBUG map low page(2c73000, 00001000)
> [    0.000000] DEBUG map low page(2c74000, 00001000)
> [    0.000000] DEBUG map low page(2c75000, 00001000)
> [    0.000000] DEBUG map low page(2c76000, 00001000)
> [    0.000000] DEBUG map low page(2c77000, 00001000)
> [    0.000000] DEBUG map low page(2c78000, 00001000)
> [    0.000000] DEBUG map low page(2c79000, 00001000)
> [    0.000000] DEBUG map low page(2c7a000, 00001000)
> [    0.000000] DEBUG map low page(2c7b000, 00001000)
> [    0.000000] DEBUG map low page(239a000, 00001000)
> [    0.000000] DEBUG map low page(239b000, 00001000)
> [    0.000000] DEBUG map low page(239c000, 00001000)
> [    0.000000] DEBUG map low page(239d000, 00001000)

these already mapped.

> [    0.000000] DEBUG alloc low page(cf93d000, 00001000)
> [    0.000000] DEBUG alloc low page(cf93e000, 00001000)
> [    0.000000] DEBUG alloc low page(cf93f000, 00001000)
> [    0.000000] DEBUG alloc low page(cf940000, 00001000)
> [    0.000000] DEBUG alloc low page(cf941000, 00001000)
> [    0.000000] DEBUG alloc low page(cf942000, 00001000)
> [    0.000000] DEBUG alloc low page(cf943000, 00001000)
> [    0.000000] DEBUG alloc low page(cf944000, 00001000)
> [    0.000000] DEBUG alloc low page(cf945000, 00001000)
> [    0.000000] DEBUG alloc low page(cf946000, 00001000)
> [    0.000000] DEBUG alloc low page(cf947000, 00001000)
> [    0.000000] DEBUG alloc low page(cf948000, 00001000)
> [    0.000000] DEBUG alloc low page(cf949000, 00001000)
> [    0.000000] DEBUG alloc low page(cf94a000, 00001000)
> [    0.000000] DEBUG alloc low page(cf94b000, 00001000)
> [    0.000000] DEBUG alloc low page(cf94c000, 00001000)
> [    0.000000] DEBUG alloc low page(cf94d000, 00001000)
> [    0.000000] DEBUG alloc low page(cf94e000, 00001000)
> [    0.000000] DEBUG alloc low page(cf94f000, 00001000)
> [    0.000000] DEBUG alloc low page(cf950000, 00001000)
> [    0.000000] DEBUG alloc low page(cf951000, 00001000)
> [    0.000000] DEBUG alloc low page(cf952000, 00001000)
> [    0.000000] DEBUG alloc low page(cf953000, 00001000)
> [    0.000000] DEBUG alloc low page(cf954000, 00001000)
> [    0.000000] DEBUG alloc low page(cf955000, 00001000)
> [    0.000000] DEBUG alloc low page(cf956000, 00001000)
> [    0.000000] DEBUG alloc low page(cf957000, 00001000)
> [    0.000000] DEBUG alloc low page(cf958000, 00001000)
> [    0.000000] DEBUG alloc low page(cf959000, 00001000)
> [    0.000000] DEBUG alloc low page(cf95a000, 00001000)
> [    0.000000] DEBUG alloc low page(cf95b000, 00001000)
> [    0.000000] DEBUG alloc low page(cf95c000, 00001000)
> [    0.000000] DEBUG alloc low page(cf95d000, 00001000)
> [    0.000000] DEBUG alloc low page(cf95e000, 00001000)
> [    0.000000] DEBUG alloc low page(cf95f000, 00001000)
> [    0.000000] DEBUG alloc low page(cf960000, 00001000)
> [    0.000000] DEBUG alloc low page(cf961000, 00001000)
> [    0.000000] DEBUG alloc low page(cf962000, 00001000)
> [    0.000000] DEBUG alloc low page(cf963000, 00001000)
> [    0.000000] DEBUG alloc low page(cf964000, 00001000)
> [    0.000000] DEBUG alloc low page(cf965000, 00001000)

these are not mapped yet, and need to use early_ioremap.

>
>
> if this is the case the introduction of map_low_page is unnecessary and
> actually makes the kernel map pages that are already mapped anyway.
> Am I correct?

initial mapping only to 512M, and new page for page tables could above that.

Yinghai
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/