Re: [PATCH]2.6.7 MSI-X Update

[Roland Dreier <roland@topspin.com>]

Long, welcome back.  I will review your patch as soon as I have a
little free time.

    Roland> 2) removing this overloaded function from free_irq() will
    Roland> also make driver code clearer and easier to maintain.

    long> I need this overloaded function to keep track of the state
    long> of each MSI/MSI-X vector. This allows me to generate a
    long> BUG_ON() if driver calls
    long> pci_disable_msi()/pci_disable_msix() without calling
    long> free_irq() for all MSI/MSI-X vector(s).

I think it's OK to keep track of which MSI/MSI-X vectors have ISRs
attached and which don't.  However I don't think free_irq() should
return a vector to the free pool (that might be used by other
devices).  In other words it should be OK for a correct driver to do

	pci_enable_msix(pdev...);
	request_irq(msix_vector1);
	free_irq(msix_vector1);
	request_irq(msix_vector1);
	free_irq(msix_vector1);
	pci_disable_msix(pdev...);

(obviously with some code in between operations).

I think we agree on this and I will read your code to find out what
you actually do but I just wanted to make my proposed interface clear.

Thanks,
  Roland