Re: [PATCH 2/2] PCI: meson: Add missing remove callback

Hi Bjorn,

> What's the user-visible effect of this?  Does it avoid an oops?
> Reduce power usage?

I don't have hardware with this controller, so I have not observed an
oops from this.

The concrete effect is that driver unbind/module unload gets the
matching teardown for the successful probe path: the DesignWare host
bridge is unregistered with dw_pcie_host_deinit(), and the PHY is
powered off with meson_pcie_power_off() before devres releases the
driver's resources.

So yes, it should reduce power usage after unbind. The host bridge
teardown is also the matching cleanup for dw_pcie_host_init(), but I
don't have hardware evidence of an oops on unbind.

> Of the 34 instances of .probe() in drivers/pci/controller/dwc/, on 12
> implement .remove(), so if this fixes a problem, I'm wondering whether
> other drivers have the same problem.

Yes, that may deserve a broader driver-by-driver audit. I only checked
Meson for this patch. Some DWC drivers implement .remove(), and several
others set .suppress_bind_attrs, which affects the manual sysfs
bind/unbind path. Before this patch, the Meson driver had neither, while
meson_pcie_probe() calls meson_pcie_power_on() and registers the host
bridge with dw_pcie_host_init().

Thanks,
Shuvam

_______________________________________________
linux-amlogic mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-amlogic