On Tue, Nov 08, 2016 at 06:35:54PM +0100, Andrew Jones wrote:
> On Tue, Nov 08, 2016 at 10:48:32AM -0500, Peter Xu wrote:
> > On Tue, Nov 08, 2016 at 01:27:31PM +0100, Alexander Gordeev wrote:
> > > On Fri, Nov 04, 2016 at 05:41:01PM +0100, Andrew Jones wrote:
> > > > > /* Scan bus look for a specific device. Only bus 0 scanned for now. */
> > > > > -pcidevaddr_t pci_find_dev(uint16_t vendor_id, uint16_t device_id)
> > > > > +int pci_find_dev(struct pci_dev *pci_dev, uint16_t vendor_id, uint16_t device_id)
> > > > > {
> > > > > pcidevaddr_t dev;
> > > > >
> > > > > - for (dev = 0; dev < 256; ++dev) {
> > > > > + for (dev = 0; dev < PCI_DEVFN_MAX; ++dev) {
> > > >
> > > > Why introduce this PCI_DEVFN_MAX define?
> > > >
> > > > > if (pci_config_readw(dev, PCI_VENDOR_ID) == vendor_id &&
> > > > > - pci_config_readw(dev, PCI_DEVICE_ID) == device_id)
> > > > > - return dev;
> > > > > + pci_config_readw(dev, PCI_DEVICE_ID) == device_id) {
> > > > > + pci_dev_init(pci_dev, dev);
> > > > > + return 0;
> > > > > + }
> > > > > }
> > > > >
> > > > > - return PCIDEVADDR_INVALID;
> > > > > + return -1;
> > > >
> > > > Why not use bool for the ret type; true=good, false=bad?
> > >
> > > Both ways look strange to me. I would leave pci_find_dev() as is (low-
> > > level) and move pci_dev_init() outside - so a usage would be i.e.:
> > >
> > > dev = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
> > > if (dev == PCIDEVADDR_INVALID)
> > > ...
> > > pci_dev_init(&pci_dev, dev);
> >
> > IMHO actually it'll be nicer with:
> >
> > struct pci_dev *pci_find_dev(uint16_t vendor_id, uint16_t device_id);
>
> Yes, I was going to suggest this too, but then...
>
> >
> > But this needs dynamic allocation of memory, which might be an
> > overkill for kvm-unit-test. So I chose to allocate the pci_dev on
> > caller stack.
>
> ...saw the value of avoiding the alloc. That said, we do have precedent
> for allocating in lib/* code already (virtio_bind), and I am currently
> reworking x86's memory management code in order to enable malloc/calloc
> use within lib code for x86. So we could go the alloc way as well...
>
> >
> > Anyway, I take the interface issue as a matter of taste.
>
> Not quite. I see Alex's reasoning. Perhaps users want to simply query
> if a device is there (pci_find_dev), but not actually init a pci_dev
> struct each time they call it. Alex's proposal keeps pci_find_dev more
> generally useful.
Yep. So in the end of the day I would kept pci_find_dev() and added
pci_dev_alloc()/pci_dev_free() helpers (not sure about naming though).
The former would have to initialize pci_dev data to be consistent
and safely used by any other PCI API that takes pci_dev pointer.
> Thanks,
> drew
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
