Re: [Qemu-devel] [PULL 48/49] i386: populate floppy drive information in DSDT

[John Snow]

On 02/08/2016 08:14 AM, Roman Kagan wrote:
> On Fri, Feb 05, 2016 at 07:25:07PM +0100, Igor Mammedov wrote:
>> On Thu, 4 Feb 2016 23:54:13 +0200
>> "Michael S. Tsirkin" <address@hidden> wrote:
>>> -static Aml *build_fdc_device_aml(void)
>>> +static Aml *build_fdinfo_aml(int idx, uint8_t type, uint8_t cylinders,
>>> +                             uint8_t heads, uint8_t sectors)
>>> +{
>>> +    Aml *dev, *fdi;
>>> +
>>> +    dev = aml_device("FLP%c", 'A' + idx);
>>> +
>>> +    aml_append(dev, aml_name_decl("_ADR", aml_int(idx)));
>>> +
>>> +    fdi = aml_package(0x10);
>>> +    aml_append(fdi, aml_int(idx));  /* Drive Number */
>>> +    aml_append(fdi,
>>> +        aml_int(cmos_get_fd_drive_type(type)));  /* Device Type */
>>> +    aml_append(fdi,
>>> +        aml_int(cylinders - 1));  /* Maximum Cylinder Number */
>> this puts uint64_t(-1) in AML i.e. cylinders == 0 and overflow happens here
>>
>> CCing Jon
> 
> I guess this is the effect of John's fdc rework.  I used to think zero
> geometry was impossible at the time this patch was developed.
> 
> I wonder if it hasn't been fixed already by
> 
>   commit fd9bdbd3459e5b9d51534f0747049bc5b6145e07
>   Author: John Snow <address@hidden>
>   Date:   Wed Feb 3 11:28:55 2016 -0500
> 
>       fdc: fix detection under Linux
>       
>       Accidentally, I removed a "feature" where empty drives had geometry
>       values applied to them, which allows seek on empty drives to work
>       "by accident," as QEMU actually tries to disallow that.
>       
>       Seeks on empty drives should work, though, but the easiest thing is to
>       restore the misfeature where empty drives have non-zero geometries
>       applied.
>       
>       Document the hack accordingly.
>       
>       [Maintainer edit]
>       
>       This fix corrects a regression introduced in d5d47efc, where
>       pick_geometry was modified such that it would not operate on empty
>       drives, and as a result if there is no diskette inserted, QEMU
>       no longer populates it with geometry bounds. As a result, seek fails
>       when QEMU denies to move the current track, but reports success anyway.
>       This can confuse the guest, leading to kernel panics in the guest.
>       
>       
>       Signed-off-by: John Snow <address@hidden>
>       Reviewed-by: Eric Blake <address@hidden>
>       Message-id: address@hidden
> 
> Roman.
> 

Yes, hopefully solved on my end. The geometry values for an empty disk
are not well defined (they certainly don't have any *meaning*) so if you
are populating tables based on an empty drive, I just hope you also have
the mechanisms needed to update said tables when the media changes.

What do the guests use these values for? Are they fixed at boot?

--js