Re: [Qemu-devel] [PATCHv2 08/11] tests: Clean up IO handling in ide-test

[David Gibson]

On Wed, Oct 19, 2016 at 04:51:41PM +0200, Laurent Vivier wrote:
> 
> 
> On 19/10/2016 16:43, Laurent Vivier wrote:
> > 
> > 
> > On 19/10/2016 14:25, David Gibson wrote:
> >> ide-test uses many explicit inb() / outb() operations for its IO, which
> >> means it's not portable to non-x86 platforms.  This cleans it up to use
> >> the libqos PCI accessors instead.
> >>
> >> Signed-off-by: David Gibson <address@hidden>
> >> ---
> >>  tests/ide-test.c | 179 
> >> ++++++++++++++++++++++++++++++++++++-------------------
> >>  1 file changed, 118 insertions(+), 61 deletions(-)
> > 
> > Could explain why you have swapped the le16_to_cpu() and cpu_to_le16()?
> > 
> > For me, they were correct.
> 
> And I have just finished testing your series on a BE host, and ide-test
> is broken:
> 
> TEST: tests/ide-test... (pid=12472)
>   /i386/ide/identify:                                                  **
> ERROR:/home/laurent/Projects/qemu/tests/ide-test.c:518:test_identify:
> assertion failed: (ret == 0)
> FAIL

Ah, thanks for testing this.

> You should not add the cpu_to_le16():
> 
>      for (i = 0; i < 256; i++) {
> -        data = inb(IDE_BASE + reg_status);
> +        data = qpci_io_readb(dev, ide_base + reg_status);
>          assert_bit_set(data, DRDY | DRQ);
>          assert_bit_clear(data, BSY | DF | ERR);
> 
> -        ((uint16_t*) buf)[i] = inw(IDE_BASE + reg_data);
> +        buf[i] = cpu_to_le16(qpci_io_readw(dev, ide_base + reg_data));
>      }

Urgh, the endianness here is doing my head in.

So, the cpu_to_le16() was supposed to counteract the implicit
conversion from LE inside qpci_io_readw.  We're reading from the data
register here, and those are usually "streaming" style, meaning that
we want byte-order preserving rather than byte-significance
preserving.

But.. the IDENTIFY command describes most of the output in terms of
(16-bit) words, meaning I guess we do want to swap from LE in order to
interpret those.  Except that the string portions seem to be encoded
strangely, which the later string_cpu_to_be16() calls are about.

In summary, I think you're right and the cpu_to_le16() shouldn't be
there.  Seems to work on a BE host, anyway.

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature