Re: [PATCH] remoteproc: Make rproc_get_by_phandle() work for clusters

[Mathieu Poirier]

Hi Bjorn,

Did you have more comments on this?  Given that we are at rc4, it
would be nice to get this to simmer in linux-next for a while.

Thanks,
Mathieu

On Wed, 4 Jan 2023 at 14:45, Mathieu Poirier <mathieu.poirier@xxxxxxxxxx> wrote:
>
> On Wed, Jan 04, 2023 at 09:56:13AM -0600, Bjorn Andersson wrote:
> > On Tue, Jan 03, 2023 at 11:48:56AM -0700, Mathieu Poirier wrote:
> > > On Tue, 27 Dec 2022 at 08:11, Bjorn Andersson <andersson@xxxxxxxxxx> wrote:
> > > >
> > > > On Wed, Dec 14, 2022 at 03:16:43PM -0700, Mathieu Poirier wrote:
> > > > > Multi-cluster remoteproc designs typically have the following DT
> > > > > declaration:
> > > > >
> > > > >       remoteproc_cluster {
> > > > >               compatible = "soc,remoteproc-cluster";
> > > > >
> > > > >                 core0: core0 {
> > > > >                       compatible = "soc,remoteproc-core"
> > > > >                         memory-region;
> > > > >                         sram;
> > > > >                 };
> > > > >
> > > > >                 core1: core1 {
> > > > >                       compatible = "soc,remoteproc-core"
> > > > >                         memory-region;
> > > > >                         sram;
> > > > >                 }
> > > > >         };
> > > > >
> > > > > A driver exists for the cluster rather than the individual cores
> > > > > themselves so that operation mode and HW specific configurations
> > > > > applicable to the cluster can be made.
> > > > >
> > > > > Because the driver exists at the cluster level and not the individual
> > > > > core level, function rproc_get_by_phandle() fails to return the
> > > > > remoteproc associated with the phandled it is called for.
> > > > >
> > > > > This patch enhances rproc_get_by_phandle() by looking for the cluster's
> > > > > driver when the driver for the immediate remoteproc's parent is not
> > > > > found.
> > > > >
> > > >
> > > > Can you please help me understand why zynqmp_r5_remoteproc_probe()
> > > > invokes devm_of_platform_populate() to create platform_device instances
> > > > for the clusters?
> > > >
> > >
> > > Platform device instances are created for the individual cores found
> > > in the cluster, following the work done on TI's K3-R5[1].
> > >
> >
> > Right, and this is a design pattern that I've been bitten by several
> > times by now. There's no real purpose of spinning up platform_devices
> > for those nodes.
> >
>
> Calling of_platform_populate() happened before my time in this subsystem.  I
> thought you were favourable to it.  Can you give one or two examples where it caused
> you grief?
>
> > > > Why can't the platform_device for the cluster be used as parent for both
> > > > remoteprocs and then the driver deal with the subnodes within the
> > > > driver?
> > > >
> > >
> > > That is exactly how things work for both K3-R5 and R5F architectures.
> > > That said, if we use the cluster's platform device as parent of the
> > > remote processors inside the cluster, function rproc_get_by_phandle()
> > > will return the first remote processor it finds with a matching parent
> > > rather than the remote processor referenced by the phandle parameter.
> > >
> >
> > I missed the fact that we don't associate either the rproc or the rproc
> > device with the of_node, but rather just rely on the fact that
> > rproc->dev->parent->of_node is typically is the handle we're looking
> > for.
> >
> > And I don't think we'll return the first instance, because
> > rproc->dev->parent->of_node will never match the instance's of_node.
> >
>
> My first suggestion was also to use the cluster's device as parent to the remote
> processors inside the cluster but it didn't work, though the exact details are
> lost in the holiday's fairy dust.  Looking more closely at the code I think you
> are correct.
>
> >
> > I think it would be cleaner to add a of_node to struct rproc and use
> > this for matching.
> >
>
> I also considered that option but decided to proceed differently because it
> duplicates of_node information that is already available and requires
> modifications to the drivers already using rproc_get_by_phandle().   Unless
> I'm missing something we would still have to call of_platform_populate() to get
> the of_node information...  And modify the parameters to rproc_alloc(), which
> cascades exponentially.
>
> > And I do suggest that we don't of_platform_populate() in the TI driver.
> > If nothing else, doing so saves ~2kb of wasted RAM...
> >
>
> And that would require a serious refactoring exercise that, in my opinion, far
> outweigh the benefits.
>
> Thanks,
> Mathieu
>
>
> > > [1]. https://elixir.bootlin.com/linux/v6.2-rc2/source/drivers/remoteproc/ti_k3_r5_remoteproc.c#L1728
> > >
> > > > Regards,
> > > > Bjorn
> > > >
> > > > > Reported-by: Ben Levinsky <ben.levinsky@xxxxxxxxxx>
> > > > > Signed-off-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
> > > > > ---
> > > > >  drivers/remoteproc/remoteproc_core.c | 28 +++++++++++++++++++++++++++-
> > > > >  1 file changed, 27 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/remoteproc/remoteproc_core.c b/drivers/remoteproc/remoteproc_core.c
> > > > > index 1cd4815a6dd1..91f82886add9 100644
> > > > > --- a/drivers/remoteproc/remoteproc_core.c
> > > > > +++ b/drivers/remoteproc/remoteproc_core.c
> > > > > @@ -33,6 +33,7 @@
> > > > >  #include <linux/idr.h>
> > > > >  #include <linux/elf.h>
> > > > >  #include <linux/crc32.h>
> > > > > +#include <linux/of_platform.h>
> > > > >  #include <linux/of_reserved_mem.h>
> > > > >  #include <linux/virtio_ids.h>
> > > > >  #include <linux/virtio_ring.h>
> > > > > @@ -2110,7 +2111,9 @@ EXPORT_SYMBOL(rproc_detach);
> > > > >  #ifdef CONFIG_OF
> > > > >  struct rproc *rproc_get_by_phandle(phandle phandle)
> > > > >  {
> > > > > +     struct platform_device *cluster_pdev;
> > > > >       struct rproc *rproc = NULL, *r;
> > > > > +     struct device_driver *driver;
> > > > >       struct device_node *np;
> > > > >
> > > > >       np = of_find_node_by_phandle(phandle);
> > > > > @@ -2121,7 +2124,30 @@ struct rproc *rproc_get_by_phandle(phandle phandle)
> > > > >       list_for_each_entry_rcu(r, &rproc_list, node) {
> > > > >               if (r->dev.parent && device_match_of_node(r->dev.parent, np)) {
> > > > >                       /* prevent underlying implementation from being removed */
> > > > > -                     if (!try_module_get(r->dev.parent->driver->owner)) {
> > > > > +
> > > > > +                     /*
> > > > > +                      * If the remoteproc's parent has a driver, the
> > > > > +                      * remoteproc is not part of a cluster and we can use
> > > > > +                      * that driver.
> > > > > +                      */
> > > > > +                     driver = r->dev.parent->driver;
> > > > > +
> > > > > +                     /*
> > > > > +                      * If the remoteproc's parent does not have a driver,
> > > > > +                      * look for the driver associated with the cluster.
> > > > > +                      */
> > > > > +                     if (!driver) {
> > > > > +                             cluster_pdev = of_find_device_by_node(np->parent);
> >
> > Doing so also has the added benefit that we don't add an implicitly
> > requirement on the rproc-device's parent being a platform_driver.
> >
> > Regards,
> > Bjorn
> >
> > > > > +                             if (!cluster_pdev) {
> > > > > +                                     dev_err(&r->dev, "can't get parent\n");
> > > > > +                                     break;
> > > > > +                             }
> > > > > +
> > > > > +                             driver = cluster_pdev->dev.driver;
> > > > > +                             put_device(&cluster_pdev->dev);
> > > > > +                     }
> > > > > +
> > > > > +                     if (!try_module_get(driver->owner)) {
> > > > >                               dev_err(&r->dev, "can't get owner\n");
> > > > >                               break;
> > > > >                       }
> > > > > --
> > > > > 2.25.1
> > > > >