Re: [Qemu-devel] Re: [PATCH v2 2/9] pcie: helper functions for pcie extended capability.

[Michael S. Tsirkin]

On Sun, Sep 19, 2010 at 01:11:21PM +0900, Isaku Yamahata wrote:
> On Wed, Sep 15, 2010 at 03:05:13PM +0200, Michael S. Tsirkin wrote:
> > On Wed, Sep 15, 2010 at 02:50:01PM +0900, Isaku Yamahata wrote:
> > > On Wed, Sep 08, 2010 at 01:31:22PM +0300, Michael S. Tsirkin wrote:
> > > > > +
> > > > > +static void pcie_notify(PCIDevice *dev, uint16_t vector,
> > > > > +                        bool trigger, int level)
> > > > > +{
> > > > > +    /* masking/masking interrupt is handled by upper layer.
> > > > > +     * i.e. msix_notify() for MSI-X
> > > > > +     *      msi_notify()  for MSI
> > > > > +     *      pci_set_irq() for INTx
> > > > > +     */
> > > > 
> > > > So this will send another interrupt when level is 0?
> > > 
> > > Yes. The condition that triggers MSI-X/MSI can be different from
> > > the one that asserts INTx as you can see it in the following code.
> > > trigger and level are set independently.
> > 
> > Looks like a bug ...
> 
> No. It can and the spec requires it. The mode of INTx and MSI is exclusive.
> I think that it's quite reasonable to assume the basic knowledge
> of express. For example
> 
> >From 6.7.3.4. Software Notification of Hot-Plug Events
> 
> > If the Port is enabled for level-triggered interrupt signaling using
> > the INTx messages, the virtualization INTx wire must be asserted whenever
> > and as long as the following conditions are satisfied:
> 
> and the list of conditions..
> 
> > If the Port is enabled for edge-triggered interrupt signaling using
> > MSI or MSI-X, an interrupt message must be sent every time the logical
> > AND of the following conditions transitions from FALSE to TRUE:
> 
> and the list of conditions.

I guess I just don't seem to be able to map the code to spec.
I don't understand what trigger and level are.

I think it would become clearer if we have two functions:
assert and dessert - instead of attempting to encode
it all as level and trigger.

Deassert would simply do nothing for msi/msix.

> yamahata