Re: [PATCH v10 0/7] Solve postboot supplier cleanup and optimize probe ordering
From: Saravana Kannan
Date: Fri Aug 30 2019 - 00:33:06 EST
On Thu, Aug 29, 2019 at 9:43 AM Rob Herring <robh+dt@xxxxxxxxxx> wrote:
> On Thu, Aug 29, 2019 at 2:46 AM Saravana Kannan <saravanak@xxxxxxxxxx> wrote:
> > Add device-links to track functional dependencies between devices
> > after they are created (but before they are probed) by looking at
> > their common DT bindings like clocks, interconnects, etc.
> > Having functional dependencies automatically added before the devices
> > are probed, provides the following benefits:
> > - Optimizes device probe order and avoids the useless work of
> > attempting probes of devices that will not probe successfully
> > (because their suppliers aren't present or haven't probed yet).
> > For example, in a commonly available mobile SoC, registering just
> > one consumer device's driver at an initcall level earlier than the
> > supplier device's driver causes 11 failed probe attempts before the
> > consumer device probes successfully. This was with a kernel with all
> > the drivers statically compiled in. This problem gets a lot worse if
> > all the drivers are loaded as modules without direct symbol
> > dependencies.
> > - Supplier devices like clock providers, interconnect providers, etc
> > need to keep the resources they provide active and at a particular
> > state(s) during boot up even if their current set of consumers don't
> > request the resource to be active. This is because the rest of the
> > consumers might not have probed yet and turning off the resource
> > before all the consumers have probed could lead to a hang or
> > undesired user experience.
> > Some frameworks (Eg: regulator) handle this today by turning off
> > "unused" resources at late_initcall_sync and hoping all the devices
> > have probed by then. This is not a valid assumption for systems with
> > loadable modules. Other frameworks (Eg: clock) just don't handle
> > this due to the lack of a clear signal for when they can turn off
> > resources. This leads to downstream hacks to handle cases like this
> > that can easily be solved in the upstream kernel.
> > By linking devices before they are probed, we give suppliers a clear
> > count of the number of dependent consumers. Once all of the
> > consumers are active, the suppliers can turn off the unused
> > resources without making assumptions about the number of consumers.
> > By default we just add device-links to track "driver presence" (probe
> > succeeded) of the supplier device. If any other functionality provided
> > by device-links are needed, it is left to the consumer/supplier
> > devices to change the link when they probe.
> > v1 -> v2:
> > - Drop patch to speed up of_find_device_by_node()
> > - Drop depends-on property and use existing bindings
> > v2 -> v3:
> > - Refactor the code to have driver core initiate the linking of devs
> > - Have driver core link consumers to supplier before it's probed
> > - Add support for drivers to edit the device links before probing
> > v3 -> v4:
> > - Tested edit_links() on system with cyclic dependency. Works.
> > - Added some checks to make sure device link isn't attempted from
> > parent device node to child device node.
> > - Added way to pause/resume sync_state callbacks across
> > of_platform_populate().
> > - Recursively parse DT node to create device links from parent to
> > suppliers of parent and all child nodes.
> > v4 -> v5:
> > - Fixed copy-pasta bugs with linked list handling
> > - Walk up the phandle reference till I find an actual device (needed
> > for regulators to work)
> > - Added support for linking devices from regulator DT bindings
> > - Tested the whole series again to make sure cyclic dependencies are
> > broken with edit_links() and regulator links are created properly.
> > v5 -> v6:
> > - Split, squashed and reordered some of the patches.
> > - Refactored the device linking code to follow the same code pattern for
> > any property.
> > v6 -> v7:
> > - No functional changes.
> > - Renamed i to index
> > - Added comment to clarify not having to check property name for every
> > index
> > - Added "matched" variable to clarify code. No functional change.
> > - Added comments to include/linux/device.h for add_links()
> > v7 -> v8:
> > - Rebased on top of linux-next to handle device link changes in 
> > v8 -> v9:
> > - Fixed kbuild test bot reported errors (docs and const)
> > v9->v10:
> > - Changes made based on reviews on LKML  and discussions at ELC 
> > - Dropped the edit_links() patch
> > - Dropped the patch that skips linking for default bus nodes
> > - 1/7: Changed from bus.add_links() to fwnode.ops.add_links()
> > - 1/7: Update device link doc
> > - 1/7: Lots of comments/fn doc updates
> > - 1/7: Renamed device_link_check_waiting_consumers() to
> > device_link_add_missing_supplier_links()
> > - 2/7: Moved DT parsing/linking code from of/platform.c to of/property.c
> Why? You'll notice that of/property.c doesn't know anything about
> platform_device (and struct device):
> $ git grep platform_device -- drivers/of/property.c
> Everything related to platform_device goes in of/platform.c.
> Everything related to struct device only goes in of/device.c. I'd be
> okay with a new file for this too.
The only platform_device related code in what got moved to
of/property.c is the call to of_find_device_by_node(). And that's
because I forgot that function returns a platform_device --- it should
really have been called of_find_plat_device_by_node() or something
similar. Outside of that, of/property.c makes sense because that's
where the fwnode ops are implemented.
As you mentioned in the other email, just searching platform_bus is
not sufficient. So I'll have to figure something out for that. Once I
do, I think the code will be fine in of/property.c as it shouldn't
have any reference to platform_device.
Thanks for catching what I missed.