Re: [Qemu-ppc] device crash test -- isa-fdc on ppc64/powernv

[John Snow]

On 10/10/2017 03:01 AM, Eduardo Otubo wrote:
> On Mon, Oct 09, 2017 at 11:53:51PM -0300, Eduardo Habkost wrote:
>> On Tue, Oct 10, 2017 at 04:41:04AM +0200, Thomas Huth wrote:
>>> On 10.10.2017 04:37, Eduardo Habkost wrote:
>>>> On Tue, Oct 10, 2017 at 04:31:40AM +0200, Thomas Huth wrote:
>>>>> On 09.10.2017 21:50, John Snow wrote:
>>>>>> Hi Thomas, I'm not particularly familiar with the various PPC flavors,
>>>>>> so please forgive me:
>>>>>>
>>>>>> Apparently isa-fdc fails to instantiate on powernv (which appears to
>>>>>> have an ISA bus) because through one way or another, we never call:
>>>>>>
>>>>>> isa_bus_dma, invoked by
>>>>>> DMA_init, invoked by, seemingly usually,
>>>>>> - i82374_realize in hw/dma/i82374.c
>>>>>>
>>>>>> by comparison, the prep board appears to use the i82378 which in turn
>>>>>> uses the 82374 which is how we get our ISA DMA initialization.
>>>>>>
>>>>>> pnv doesn't seem to have a DMA device on its ISA bus.
>>>>>>
>>>>>> Should the powernv board be initializing something like this, or is it
>>>>>> more correct to just disable the isa-fdc device for this board for now
>>>>>> and be done with it?
>>>>>
>>>>> I don't know the powernv board very well, but maybe Cédric could answer
>>>>> that question.
>>>>>
>>>>> For reference, I assume you're talking about this crash here:
>>>>>
>>>>> qemu-system-ppc64 -M powernv -device isa-fdc
>>>>> qemu-system-ppc64: hw/block/fdc.c:2703: isabus_fdc_realize: Assertion
>>>>> `fdctrl->dma' failed.
>>>>> Aborted (core dumped)
>>>>>
>>>>> ?
>>>>>
>>>>> IIRC Eduardo also posted a patch for that one already, not sure about
>>>>> the current status, though.
>>>>
>>>> Eduardo Otubo posted a patch for a different case: assertion
>>>> failure inside DMA_init() when another device already called
>>>> DMA_init() (e.g. if two i82374 devices are instantiated).  The
>>>> fix was to make i82374 non-user-creatable.
>>>>
>>>
>>> You mean this one here:
>>>
>>> https://lists.gnu.org/archive/html/qemu-devel/2017-09/msg04100.html
>>
>> Yes.
> 

Ah, right, I knew this problem seemed familiar but I was having finding
the patch. (You never replied to my question!)

> I believe Philippe Mathieu-Daudé is taking over this since he's already 
> chaning
> DMA_init prototype. But well, if we don't have any work on this in the 
> following
> days I can do it, no problem.
> 

PMD, is this fix in your queue?


John