Re: [Qemu-devel] [PATCH 1/2] Keep track of ISA ports ISA device is using in qdev.

[Gleb Natapov <gleb@redhat.com>]

On Tue, Oct 26, 2010 at 11:49:26AM +0200, Markus Armbruster wrote:
> Gleb Natapov <gleb@redhat.com> writes:
> 
> > On Mon, Oct 25, 2010 at 08:01:13PM +0200, Markus Armbruster wrote:
> >> Gleb Natapov <gleb@redhat.com> writes:
> >> 
> >> > On Mon, Oct 25, 2010 at 05:06:51PM +0200, Markus Armbruster wrote:
> >> >> Gleb Natapov <gleb@redhat.com> writes:
> >> >> 
> >> >> > Prevent two devices from claiming the same io port.
> >> >> 
> >> >> Really?
> >> >> 
> >> >> Your new check for double-claim is in the new isa_init_ioport(), which
> >> >> is for ISADevice only.  Thus, only qdevified ISA devices can opt for
> >> >> this protection, by calling isa_init_ioport().  It doesn't protect from
> >> >> devices who don't or can't opt in, such as PCI devices.
> >> >> 
> >> > I didn't claim different. This obviously works only for ISA qdev
> >> > devices.
> >> 
> >> I read "Prevent two devices from claiming the same io port" as such.
> >> 
> > And have you read subject with that?
> 
> Yes.  Nevertheless, I found it a bit misleading, so I commented.
> 
Well, I hope it is clear now.

> >> >> Anyway, we already check for double-claim in
> >> >> register_ioport_{read,write}().  The check has issues --- hw_error() is
> >> >> wrong there for hot plug.  But it's where the check should be, isn't it?
> >> > You don't like double-claim checking? Forget about it (all 3 lines
> >> > of code). The real point of the patch is to have ISA resources used
> >> > by devices to be stored in common place (ISADevice) which allows
> >> > get_dev_path() to be implemented.
> >> 
> >> We already track I/O port use, in ioport.c.  I don't like that
> >> duplicated.  Even less so if the dupe catches fewer double-claims than
> >> the original.
> > Consider it removed although we do keep track of irqs there and this
> > tracking is also incomplete. Why is it there?
> 
> Where's "there"?
> 
In ISADevice. Look for "isabus->assigned" in isa-bus.c.

> Let's not add to the code duplication if we can help it.  It's bad
> enough as it is.
> 
> >> Perhaps your new function should wrap around register_ioport_*() instead
> >> of supplementing it.
> > register_ioport_*() is disconnected from qdev in general and from ISADEvice
> > in particular, so how "wrap around register_ioport_*()" will help me to
> > have get_dev_path() for ISABus is beyond my understanding.
> 
> I was too terse, sorry.
> 
> Maybe we should have functions for ISADevices to register ISA resources.
> Functions that are *not* disconnected from qdev.  Instead of code like
> 
>         register_ioport_write(pm_io_base, 64, 2, pm_ioport_writew, s);
>         register_ioport_read(pm_io_base, 64, 2, pm_ioport_readw, s);
>         register_ioport_write(pm_io_base, 64, 4, pm_ioport_writel, s);
>         register_ioport_read(pm_io_base, 64, 4, pm_ioport_readl, s);
>         isa_init_ioport_range(dev, pm_ioport_base, 64);
> 
> or, with Avi's proposed interface
> 
>         ioport_register(&s->ioport, pm_io_base, 64);
>         isa_init_ioport_range(dev, pm_ioport_base, 64);
> 
> we'd get something like
> 
>         isa_ioport_register(dev, &s->ioport, pm_io_base, 64);
> 
> Isn't that neater?
> 
Neater, but out of scope for my work. Obviously all resource
management should go via qdev at the end.

> Since some PCI devices also register ISA resources, we'd have to 
> offer functions for them as well, properly integrated.
> 
> Now, I didn't mean to ask you to do all that as a precondition for
> getting your patch in.  I was just observing.
OK.


--
			Gleb.