Re: [PATCH] cxl/port: Fix port to pci device assumptions in read_cdat_data()

[Dan Williams <dan.j.williams@intel.com>]

Lukas Wunner wrote:
> On Sat, Apr 22, 2023 at 01:54:00PM -0700, Dan Williams wrote:
> > Lukas Wunner wrote:
> > > On Sat, Apr 22, 2023 at 10:35:02AM +0200, Lukas Wunner wrote:
> > > > I note however that before af0a6c3587dc, xa_for_each() was run on an
> > > > xarray which was not initialized with xa_init() on non-pci cxl ports.
> > > > (xa_init() was run from cxl_pci_probe() -> devm_cxl_pci_create_doe()
> > > > but xa_for_each() was run from read_cdat_data() -> find_cdat_doe()
> > > > for non-pci cxl ports as well.)
> > > > 
> > > > Hence can't this crash prior to af0a6c3587dc as well?
> > > 
> > > After taking another look with a fresh pair of eyeballs I think
> > > you may see splats with CONFIG_DEBUG_LOCK_ALLOC=y (even prior to
> > > af0a6c3587dc on non-pci cxl devices) due to the missing xa_init().
> > 
> > The xa_init() was included before in the removed call to
> > devm_cxl_pci_create_doe().
> 
> ... which was called from cxl_pci_probe(), so only if a pci_dev exists?
> 
> I guess I must be missing something...

Oh, no you're right in the sense that the code will use an
only-zero-initialized xarray before the move to the PCI core, but a zero
initialized xarray is sufficient. Recall that xarray walks are only RCU
protected, so the xarray spinlock is not used for xa_for_each().

The change that makes this a failure is now the doe_mbs xarray is not
even allocated in the !pci_dev case.