Re: [RFC PATCH 1/6] Revert "PCI: Fix is_added/is_busmaster race condition"

[Benjamin Herrenschmidt]

On Mon, 2018-08-20 at 09:17 +0200, Lukas Wunner wrote:
> On Mon, Aug 20, 2018 at 12:10:59PM +1000, Benjamin Herrenschmidt wrote:
> > I chose to create a new mutex which we should be able to address other
> > similar races if we find them. The other solutions that I dismissed
> > were:
> > 
> >  - Using the device_lock. A explained previously, this is tricky, I
> > prefer not using this for anything other than locking against
> > concurrent add/remove. The main issue is that drivers will be sometimes
> > called in context where that's already held, so we can't take it inside
> > pci_enable_device() and I'd rather not add new constraints such as
> > "pci_enable_device() must be only called from probe() unless you also
> > take the device lock". It would be tricky to audit everybody...
> > 
> >  - Using a global mutex. We could move the bridge lock from AER to core
> > code for example, and use that. But it doesn't buy us much, and
> > slightly redecuces parallelism. It also makes it a little bit more
> > messy to walk up the bridge chain, we'd have to do a
> > pci_enable_device_unlocked or something, messy.
> 
> +1 from my side for adding a struct mutex to struct pci_dev to protect
> state changes.

Ok thanks. This is what my patch proposes. We can use it later to
protect more things if we wish to do so.

> The device_lock() primarily protects binding / unbinding of the device
> and pci_dev state may have to be changed while binding / unbinding.

Yup, precisely.

> A global lock invites deadlocks if multiple devices are added / removed
> concurrently where one is a parent of the other.  (Think hot-removal of
> multiple devices on a Thunderbolt daisy-chain.)

Yes.

> As said I'd also welcome folding PCI_DEV_DISCONNECTED into enum
> pci_channel_state, either as an additional state or by using
> pci_channel_io_perm_failure.

Ok. I have that in my tentative series but I think for robustness, I
should make the error_state field atomically updated in order to ensure
that no transition out of "disconnected" can happen while racing with
concurrent error_state updates at interrupt time (at least with EEH, it
can be updated from any read{b,w,l,q}).

I'll do a bit more work on the patches this week as time permits and
send a non-RFC series.

Cheers,
Ben. 

> Thanks,
> 
> Lukas