Re: [Qemu-devel] [Qemu-ppc] [PATCH v3 2/4] ppc64-dump: Support dump for little endian ppc64

[Greg Kurz]

On Wed, 07 May 2014 15:54:10 -0500
Tom Musta <address@hidden> wrote:

> On 5/7/2014 2:02 PM, Tom Musta wrote:
> 
> > It feels like there is an endianness issue here but I have not yet been able
> > to put my finger on it.
> 
> OK ... after more thought and scribbling ... here is what I mean ....
> 
> Suppose I have a 64-bit value 0xa0a1a2a3a4a5a6a7 stored in guest memory and it
> gets loaded into a GPR.  If I follow the dump code and view all four 
> combinations
> of guest/host big/little endian, I convince myself that the big endian guest
> code writes the byte sequence 0xa0, 0xa1, 0xa2, ..., 0xa7 into the file.  And
> the little endian guest dumps contain the sequence 0xa7, 0xa6, ..., 0xa0.
> 
> This make sense ... the endianness indicated in the dump header and the 
> endianness of
> the dump file layout are consistent, irrespective of the host endianness.
> 
> If I take this a step further and consider a 128-bit value 
> 0xa0a1a2a3a4a5a6a7a8a9aaabacadaeaf
> stored in guest memory and look at the AVR structure (via printf or debugger) 
> after doing
> a 128-bit lvx load, I get the following:
> 
>   +------+-------+------------------+------------------+---------------------+
>   | Host | Guest | avr.u64[0]       | avr.u64[1]       | file sequence       |
>   +------+-------+------------------+------------------+---------------------+
>   | BE   | BE    | a0a1a2a3a4a5a6a7 | a8a9aaabacadaeaf | a0,...,a7,a8,...,af |
>   | LE   | BE    | a8a9aaabacadaeaf | a0a1a2a3a4a5a6a7 | a8,...,af,a0,...,a7 |
>   | BE   | LE    | a0a1a2a3a4a5a6a7 | a8a9aaabacadaeaf | a7,...,a0,af,...,a8 |
>   | LE   | LE    | a8a9aaabacadaeaf | a0a1a2a3a4a5a6a7 | af,...,a8,a7,...,a0 |
>   +------+-------+------------------+------------------+---------------------+
> 
> The last column represents how I think the proposed dump code will write bytes
> to disk.  Notice that if you invert the (64-bit) array elements, then the two
> BE dumps look alike and the two LE dumps look alike.  If you swap array u64
> elements on LE hosts, and also swap on LE guests, then you get a byte sequence
> that looks like a 128-bit integer in all cases.
> 
> 

Then we already have an issue with the current code...

-- 
Gregory Kurz                                     address@hidden
                                                 address@hidden
Software Engineer @ IBM/Meiosys                  http://www.ibm.com
Tel +33 (0)562 165 496

"Anarchy is about taking complete responsibility for yourself."
        Alan Moore.