[dpdk-dev] [PATCH 2/2] virtio: fix PCI config err handling

[Luca Boccassi]

On Tue, 2018-08-21 at 10:40 +0800, Tiwei Bie wrote:
> On Mon, Aug 20, 2018 at 05:45:35PM +0100, Luca Boccassi wrote:
> > On Mon, 2018-08-20 at 16:18 +0800, Tiwei Bie wrote:
> > > On Thu, Aug 16, 2018 at 07:49:43PM +0100, Luca Boccassi wrote:
> > > > On Thu, 2018-08-16 at 19:47 +0100, Luca Boccassi wrote:
> > > > > From: Brian Russell <brussell at brocade.com>
> > > > > 
> > > > > In virtio_read_caps and vtpci_msix_detect,
> > > > > rte_pci_read_config
> > > > > returns
> > > > > the number of bytes read from PCI config or < 0 on error.
> > > > > If less than the expected number of bytes are read then log
> > > > > the
> > > > > failure and return rather than carrying on with garbage.
> > > > > 
> > > > > Fixes: 6ba1f63b5ab0 ("virtio: support specification 1.0")
> > > > > 
> > > > > Signed-off-by: Brian Russell <brussell at brocade.com>
> > > > > Signed-off-by: Luca Boccassi <bluca at debian.org>
> > > > > ---
> > > > > v2: handle additional rte_pci_read_config incomplete reads
> > > > > 
> > > > >  drivers/net/virtio/virtio_pci.c | 35 +++++++++++++++++++++
> > > > > ----
> > > > > ----
> > > > > ----
> > > > >  1 file changed, 22 insertions(+), 13 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/net/virtio/virtio_pci.c
> > > > > b/drivers/net/virtio/virtio_pci.c
> > > > > index 6bd22e54a6..ff6f96f361 100644
> > > > > --- a/drivers/net/virtio/virtio_pci.c
> > > > > +++ b/drivers/net/virtio/virtio_pci.c
> > > > 
> > > > ...
> > > > > @@ -610,9 +613,13 @@ virtio_read_caps(struct rte_pci_device
> > > > > *dev,
> > > > > struct virtio_hw *hw)
> > > > >  			hw->common_cfg = get_cfg_addr(dev,
> > > > > &cap);
> > > > >  			break;
> > > > >  		case VIRTIO_PCI_CAP_NOTIFY_CFG:
> > > > > -			rte_pci_read_config(dev, &hw-
> > > > > > notify_off_multiplier,
> > > > > 
> > > > > -					4, pos +
> > > > > sizeof(cap));
> > > > > -			hw->notify_base = get_cfg_addr(dev,
> > > > > &cap);
> > > > > +			/* Avoid half-reads into hw */
> > > > > +			ret = rte_pci_read_config(dev,
> > > > > &multiplier,
> > > > > 4,
> > > > > +					pos + sizeof(cap));
> > > > > +			if (ret == 4) {
> > > > > +				hw->notify_off_multiplier =
> > > > > multiplier;
> > > > > +				hw->notify_base =
> > > > > get_cfg_addr(dev,
> > > > > &cap);
> > > > > +			}
> > > > >  			break;
> > > > >  		case VIRTIO_PCI_CAP_DEVICE_CFG:
> > > > >  			hw->dev_cfg = get_cfg_addr(dev,
> > > > > &cap);
> > > > 
> > > > Tiwei: not 100% sure what's the best way to handle a failure
> > > > here,
> > > > this
> > > > will avoid writing to *hw in case of errors. Let me know if
> > > > this is
> > > > OK.
> > > 
> > > My concern is about reading the virtio capability directly.
> > > With this patch, it will give up reading other capabilities
> > > when failed to read a full virtio PCI capability structure
> > > (the returned bytes are less than the expected bytes) even
> > > though when the capability it's reading isn't a virtio vendor
> > > specific capability. I'm not quite sure whether it will bring
> > > any bad consequence. How about just reading the first two
> > > bytes first? Something like this:
> > > 
> > > https://github.com/freebsd/freebsd/blob/72445a41b3ff/sys/dev/pci/
> > > pci.
> > > c#L1491-L1497
> > 
> > I'm not sure, to be honest. That bit didn't give me trouble at all,
> > so
> > at this point I'd much rather leave it alone so that the
> > maintainers
> > can take care of it how they see fit, if necessary :-)
> > 
> > I've sent a v3 that removes that individual change.
> 
> My concern isn't about the above change (which handled the
> errors when reading multiplier). In fact, above change looks
> good to me! :-)  I mean the below changes in this patch:
> 
>  	while (pos) {
>  		ret = rte_pci_read_config(dev, &cap, sizeof(cap),
> pos);
> -		if (ret < 0) {
> -			PMD_INIT_LOG(ERR,
> -				"failed to read pci cap at pos: %x",
> pos);
> +		if (ret != sizeof(cap)) {
> +			PMD_INIT_LOG(DEBUG,
> +				     "failed to read pci cap at pos:
> %x ret %d",
> +				     pos, ret);
>  			break;
>  		}
> 
> With this change, it will give up reading other capabilities
> when failed to read a full virtio PCI capability structure
> (the returned bytes are less than the expected bytes) even
> though when the capability it's reading isn't a virtio vendor
> specific capability. Maybe it would be better to read the
> first two bytes first and check whether it's the capability
> we want to parse (i.e. vendor capability and MSIX capability).
> Something like this:
> 
> https://github.com/freebsd/freebsd/blob/72445a41b3ff/sys/dev/pci/pci.
> c#L1491-L1497
> 
> How do you think?
> 
> Thanks

Ah I see what you mean now, sorry. Would something like the following
be what you are looking for:

		ret = rte_pci_read_config(dev, &cap, 2, pos);
		if (ret != 2) {
			PMD_INIT_LOG(DEBUG,
				     "failed to read pci cap at pos: %x ret %d",
				     pos, ret);
			break;
		}
		if (cap.cap_vndr != PCI_CAP_ID_MSIX &&
				cap.cap_vndr != PCI_CAP_ID_VNDR) {
			goto next;
		}

		ret = rte_pci_read_config(dev, &cap, sizeof(cap), pos);
		if (ret != sizeof(cap)) {
			PMD_INIT_LOG(DEBUG,
				     "failed to read pci cap at pos: %x ret %d",
				     pos, ret);
			break;
		}

-- 
Kind regards,
Luca Boccassi