Re: [Qemu-devel] [PATCH] dma/i82374: avoid double creation of i82374 device

[Eduardo Habkost <ehabkost@redhat.com>]

On Fri, Sep 01, 2017 at 05:34:34PM +0200, Markus Armbruster wrote:
> Eduardo Habkost <ehabkost@redhat.com> writes:
> 
> > i82374 is compiled in only on ppc and sh4, so I'm CCing the
> > maintainers for those architectures.
> >
> > On Fri, Sep 01, 2017 at 01:03:32PM +0200, Eduardo Otubo wrote:
> >> When used with the following command line:
> >> 
> >>  ./ppc64-softmmu/qemu-system-ppc64 -S -machine 40p,accel=tcg -device i82374
> >> 
> >> QEMU with machine type 40p already creates the device i82374. If
> >> specified in the command line, it will try to create it again, hence
> >> generating the error.
> >
> > Which error?
> >
> >
> >>                       One way to avoid this problem is to set
> >> user_creatable=false.
> >> 
> >> Signed-off-by: Eduardo Otubo <otubo@redhat.com>
> >
> > The patch does more than just avoiding double creation: it
> > prevents usage of "-device i82374" completely.
> >
> > Maybe nobody needs it to work with -device today (would the
> > device even work?) and it is OK to set user_creatable=false until
> > we fix the crash.  But we need to be sure of that.
> >
> >> ---
> >>  hw/dma/i82374.c | 1 +
> >>  1 file changed, 1 insertion(+)
> >> 
> >> diff --git a/hw/dma/i82374.c b/hw/dma/i82374.c
> >> index 6c0f975df0..5275d822e0 100644
> >> --- a/hw/dma/i82374.c
> >> +++ b/hw/dma/i82374.c
> >> @@ -139,6 +139,7 @@ static void i82374_class_init(ObjectClass *klass, void *data)
> >>      dc->realize = i82374_realize;
> >>      dc->vmsd = &vmstate_i82374;
> >>      dc->props = i82374_properties;
> >> +    dc->user_creatable = false;
> >
> > A "Reason:" comment explaining why user_creatable=false is
> > mandatory.  See the comment above user_creatable declaration in
> > qdev-core.h for reference.
> >
> > I suggest the following:
> >
> >     /*
> >      * Reason: i82374_realize() crashes (assertion failure inside isa_bus_dma()
> >      *         if the device is instantiated twice.
> >      */
> 
> We need to find out *why* it crashes.  Once we know, we can likely write
> a better comment.

It crashes because isa_bus_dma() isn't supposed to be called
twice for the same bus.

Making isa_bus_dma()/DMA_init()/i82374_realize() return an error
instead of asserting would be even better than setting
user_creatable=false.

-- 
Eduardo