Re: [PATCH] PCI/DPC: Add Software Trigger as reset method

[Bjorn Helgaas <helgaas@kernel.org>]

Hi Lukas,

On Tue, Nov 29, 2022 at 08:35:55AM +0100, Lukas Wunner wrote:
> Add DPC Software Trigger as a reset method to be used for silicon
> validation among other things:
> 
>   # echo dpc_sw_trigger > reset_method
>   # echo 1 > reset
> 
> After validating DPC, the default reset_method(s) may be reinstated:
> 
>   # echo default > reset_method
> 
> Writing the DPC Control Register requires that control was granted by
> firmware, so expose the reset_method only if DPC is native.  (And AER,
> which must always be granted or denied in unison per PCI Firmware Spec
> r3.3 table 4-5.)
> 
> The reset attribute in sysfs is meant to reset a single PCI Function,
> but DPC resets the entire hierarchy below the parent.  So only expose
> the reset method on PCI Functions without siblings or children.
> Checking for that may happen both *before* the PCI Function has been
> added to the bus list (via pci_device_add() -> pci_init_capabilities())
> and *after* (via reset_method_store()), hence differentiate between
> those two cases on reset probing.
> 
> It would be useful for silicon validation to have a separate sysfs
> attribute for PCI bridges to reset their downstream hierarchy.  Prepare
> for introduction of such an attribute by adding separate functions
> pci_dpc_sw_trigger() (to reset the hierarchy below a bridge) and
> pci_dpc_sw_trigger_parent() (to reset a single PCI Function), where the
> latter calls the former to trigger DPC on the parent bridge.

Does this add functionality above what the "bus" method (Secondary Bus
Reset) provides?  If it's effectively the same as SBR, I'm not sure
it's worth complicating the user interface just to support silicon
validation.

Bjorn