[PATCH v2] ARM: mm: support big-endian page tables

[Jianguo Wu]

On 2014/4/14 19:14, Marc Zyngier wrote:

> On 14/04/14 11:43, Will Deacon wrote:
>> (catching up on old email)
>>
>> On Tue, Mar 18, 2014 at 07:35:59AM +0000, Jianguo Wu wrote:
>>> Cloud you please take a look at this?
>>
>> [...]
>>
>>> On 2014/2/17 15:05, Jianguo Wu wrote:
>>>> When enable LPAE and big-endian in a hisilicon board, while specify
>>>> mem=384M mem=512M at 7680M, will get bad page state:
>>>>
>>>> Freeing unused kernel memory: 180K (c0466000 - c0493000)
>>>> BUG: Bad page state in process init  pfn:fa442
>>>> page:c7749840 count:0 mapcount:-1 mapping:  (null) index:0x0
>>>> page flags: 0x40000400(reserved)
>>>> Modules linked in:
>>>> CPU: 0 PID: 1 Comm: init Not tainted 3.10.27+ #66
>>>> [<c000f5f0>] (unwind_backtrace+0x0/0x11c) from [<c000cbc4>] (show_stack+0x10/0x14)
>>>> [<c000cbc4>] (show_stack+0x10/0x14) from [<c009e448>] (bad_page+0xd4/0x104)
>>>> [<c009e448>] (bad_page+0xd4/0x104) from [<c009e520>] (free_pages_prepare+0xa8/0x14c)
>>>> [<c009e520>] (free_pages_prepare+0xa8/0x14c) from [<c009f8ec>] (free_hot_cold_page+0x18/0xf0)
>>>> [<c009f8ec>] (free_hot_cold_page+0x18/0xf0) from [<c00b5444>] (handle_pte_fault+0xcf4/0xdc8)
>>>> [<c00b5444>] (handle_pte_fault+0xcf4/0xdc8) from [<c00b6458>] (handle_mm_fault+0xf4/0x120)
>>>> [<c00b6458>] (handle_mm_fault+0xf4/0x120) from [<c0013754>] (do_page_fault+0xfc/0x354)
>>>> [<c0013754>] (do_page_fault+0xfc/0x354) from [<c0008400>] (do_DataAbort+0x2c/0x90)
>>>> [<c0008400>] (do_DataAbort+0x2c/0x90) from [<c0008fb4>] (__dabt_usr+0x34/0x40)
>>
>> [...]
>>
>>>> The bug is happened in cpu_v7_set_pte_ext(ptep, pte):
>>>> when pte is 64-bit, for little-endian, will store low 32-bit in r2,
>>>> high 32-bit in r3; for big-endian, will store low 32-bit in r3,
>>>> high 32-bit in r2, this will cause wrong pfn stored in pte,
>>>> so we should exchange r2 and r3 for big-endian.
>>

Hi Marc,
How about this:

The bug is happened in cpu_v7_set_pte_ext(ptep, pte):
- It tests the L_PTE_NONE in one word on the other, and possibly clear L_PTE_VALID
  tst	r3, #1 << (57 - 32)		@ L_PTE_NONE
  bicne	r2, #L_PTE_VALID
- Same for L_PTE_DIRTY, respectively setting L_PTE_RDONLY

As for LPAE, the pte is 64-bits, and the value of r2/r3 is depending on the endianness,
for little-endian, will store low 32-bit in r2, high 32-bit in r3,
for big-endian, will store low 32-bit in r3, high 32-bit in r2, 
this will cause wrong bit is cleared or set, and get wrong pfn.
So we should exchange r2 and r3 for big-endian.

Thanks,
Jianguo Wu.

>> I believe that Marc (added to CC) has been running LPAE-enabled, big-endian
>> KVM guests without any issues, so it seems unlikely that we're storing the
>> PTEs backwards. Can you check the configuration of SCTLR.EE?
> 
> So, for the record:
> 
> root at when-the-lie-s-so-big:~# cat /proc/cpuinfo 
> processor	: 0
> model name	: ARMv7 Processor rev 4 (v7b)
> Features	: swp half thumb fastmult vfp edsp thumbee neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm 
> CPU implementer	: 0x41
> CPU architecture: 7
> CPU variant	: 0x0
> CPU part	: 0xc07
> CPU revision	: 4
> 
> processor	: 1
> model name	: ARMv7 Processor rev 4 (v7b)
> Features	: swp half thumb fastmult vfp edsp thumbee neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm 
> CPU implementer	: 0x41
> CPU architecture: 7
> CPU variant	: 0x0
> CPU part	: 0xc07
> CPU revision	: 4
> 
> Hardware	: Dummy Virtual Machine
> Revision	: 0000
> Serial		: 0000000000000000
> root at when-the-lie-s-so-big:~# uname -a
> Linux when-the-lie-s-so-big 3.14.0+ #2465 SMP PREEMPT Tue Apr 8 13:05:11 BST 2014 armv7b GNU/Linux
> 
> Now, looking at the patch, I think it makes some sense:
> - Depending on the endianness, we have to test the L_PTE_NONE in one 
> word on the other, and possibly clear L_PTE_VALID
> - Same for L_PTE_DIRTY, respectively setting L_PTE_RDONLY
> 
> The commit message looks wrong though, as it mention the PTE storage in 
> memory (which looks completely fine to me, and explain why I was able to
> boot a guest). As none of my guest RAM is above 4GB IPA, I didn't see 
> the corruption of bit 32 in the PTE (which should have been bit 0,
> corresponding to L_PTE_VALID).
> 
> So, provided that the commit message is rewritten to match the what it does,
> I'm fine with that patch.
> 
> Thanks,
> 
> 	M.