Re: [PATCH] hw/ide: Don't divide by zero if guest specifies 0 sectors

I have spent some time on what the ATA contract actually says about
INITIALIZE DEVICE PARAMETERS, mostly because of the reasoning quoted
below, and ended up with a set of findings.

On 30.06.2026 12:42, Peter Maydell wrote:
> INITIALIZE DRIVE
> PARAMETERS is specified to not check the input values for validity;
> instead no error is posted until some other command makes an illegal
> access. So we have to cope with a 0 divisor here.

1) That premise does not hold for ATA-2 and later. ATA-5 8.16.8, with
   the same wording in ATA-4 8.16.8:

     If the host requests a CHS translation that is not supported by
     the device, the device shall return command aborted. [...]
     If the requested CHS translation is not supported, the device
     shall fail all media access commands with an ID Not Found error
     until a valid CHS translation is established.

   ATA-2 7.18 speaks to the "no error is posted" behaviour directly in
   NOTE 26: previous ATA specifications were unclear, some
   implementations indicate no error at all, but most of those still
   fail media access commands. The command was dropped entirely in
   ATA-6 along with CHS addressing.

2) Zero sectors per logical track is an unsupported translation by
   definition, not a borderline case: ATA-5 6.2 numbers CHS sectors
   from one, and ATA-2 D.2.8 limits IDENTIFY DEVICE word 56 to values
   between 1 and 255. So 8.16.6 has us abort the command, and the
   error belongs there rather than on the later access.

Your question below the --- line, whether we should check the addresses
a guest uses against the CHS limits it selected, is a real gap as well,
but I would keep it apart from the crash fix.

For me this changes where the fix has to live. There are further
consequences, one of them around migration, but they are much easier to
discuss with the code in hand, so let me send that first.

Den