| Message ID | AANLkTimsKcn3n90cyqQTQ4Ycy2dlsHdi7afeiNX4M0v5@mail.gmail.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Delegated to: | Grant Likely |
| Headers | show |
Hello, On 7/22/10, Grant Likely <grant.likely@secretlab.ca> wrote: > On Thu, Jul 22, 2010 at 10:48 AM, Dmitry Eremin-Solenikov > <dbaryshkov@gmail.com> wrote: >> Hello, >> >> On Thu, Jul 22, 2010 at 7:38 PM, Kumar Gala <galak@kernel.crashing.org> >> wrote: >>> >>> On Jul 21, 2010, at 7:03 PM, Dmitry Eremin-Solenikov wrote: >>> >>>> Currently (as mpc8540-pci) devices are not created on of_platform bus, >>>> mpc85xx_edac can't probe to them. Follow the change to dts trees to bind >>>> not to the main mpc8540-pci node but to special mpc85xx-pci-error nodes, >>>> present on soc bus. >>>> >>>> Signed-off-by: Dmitry Eremin-Solenikov <dbaryshkov@gmail.com> >>>> --- >>>> drivers/edac/mpc85xx_edac.c | 18 +++++++++--------- >>>> 1 files changed, 9 insertions(+), 9 deletions(-) >>> >>> Nak. >>> >>> We already have a node in the dts for the PCI controller. Lets update >>> the platform code to add the pci controller to the of_platform_bus_probe >>> list. >> >> I've had that idea. However it's really look strange to me to call >> of_platform_bus_probe() on the bus node, for which we (IMO) explicitly >> won't like for >> child devices (PCI devices) to be added to of_platform bus. Would it >> be suitable to just call of_platform_device_create for it (Or do i >> miss someth<ing)? > > Try the attached patch (lightly tested). If it works for you then > I'll post it for wider review. Yes, this patch worked for me. However it looks a bit like a hack for me.
On Fri, Jul 23, 2010 at 6:20 PM, Dmitry Eremin-Solenikov <dbaryshkov@gmail.com> wrote: > Hello, > > On 7/22/10, Grant Likely <grant.likely@secretlab.ca> wrote: >> On Thu, Jul 22, 2010 at 10:48 AM, Dmitry Eremin-Solenikov >> <dbaryshkov@gmail.com> wrote: >>> Hello, >>> >>> On Thu, Jul 22, 2010 at 7:38 PM, Kumar Gala <galak@kernel.crashing.org> >>> wrote: >>>> >>>> On Jul 21, 2010, at 7:03 PM, Dmitry Eremin-Solenikov wrote: >>>> >>>>> Currently (as mpc8540-pci) devices are not created on of_platform bus, >>>>> mpc85xx_edac can't probe to them. Follow the change to dts trees to bind >>>>> not to the main mpc8540-pci node but to special mpc85xx-pci-error nodes, >>>>> present on soc bus. >>>>> >>>>> Signed-off-by: Dmitry Eremin-Solenikov <dbaryshkov@gmail.com> >>>>> --- >>>>> drivers/edac/mpc85xx_edac.c | 18 +++++++++--------- >>>>> 1 files changed, 9 insertions(+), 9 deletions(-) >>>> >>>> Nak. >>>> >>>> We already have a node in the dts for the PCI controller. Lets update >>>> the platform code to add the pci controller to the of_platform_bus_probe >>>> list. >>> >>> I've had that idea. However it's really look strange to me to call >>> of_platform_bus_probe() on the bus node, for which we (IMO) explicitly >>> won't like for >>> child devices (PCI devices) to be added to of_platform bus. Would it >>> be suitable to just call of_platform_device_create for it (Or do i >>> miss someth<ing)? >> >> Try the attached patch (lightly tested). If it works for you then >> I'll post it for wider review. > > Yes, this patch worked for me. However it looks a bit like a hack for me. I'll probably refine it a bit before merging, but I don't think it is a hack. It reflects the behaviour that makes sense when registering devices hanging off the root node. If a device node is a child of the root, then we know it isn't hanging off an i2c or pci bus, or anything else. It is essentially a system device. The troublesome bit is that the root node also has memory, cpus, chosen and aliases nodes which are not devices. In the vast majority of cases, we want all the device nodes that are children of the root to be registered, but we don't want to register the special nodes. Checking for the presence of a compatible property is a pretty good test for determining whether or not a node actually represents a device, especially because all users of of_platform_bus_probe() seem to be FDT users where we've been very strict about enforcing that drivers must use the compatible property for matching to device nodes. Cheers, g.
On 7/24/10, Grant Likely <grant.likely@secretlab.ca> wrote: > On Fri, Jul 23, 2010 at 6:20 PM, Dmitry Eremin-Solenikov > <dbaryshkov@gmail.com> wrote: >> Hello, >> >> On 7/22/10, Grant Likely <grant.likely@secretlab.ca> wrote: >>> On Thu, Jul 22, 2010 at 10:48 AM, Dmitry Eremin-Solenikov >>> <dbaryshkov@gmail.com> wrote: >>>> Hello, >>>> >>>> On Thu, Jul 22, 2010 at 7:38 PM, Kumar Gala <galak@kernel.crashing.org> >>>> wrote: >>>>> >>>>> On Jul 21, 2010, at 7:03 PM, Dmitry Eremin-Solenikov wrote: >>>>> >>>>>> Currently (as mpc8540-pci) devices are not created on of_platform bus, >>>>>> mpc85xx_edac can't probe to them. Follow the change to dts trees to >>>>>> bind >>>>>> not to the main mpc8540-pci node but to special mpc85xx-pci-error >>>>>> nodes, >>>>>> present on soc bus. >>>>>> >>>>>> Signed-off-by: Dmitry Eremin-Solenikov <dbaryshkov@gmail.com> >>>>>> --- >>>>>> drivers/edac/mpc85xx_edac.c | 18 +++++++++--------- >>>>>> 1 files changed, 9 insertions(+), 9 deletions(-) >>>>> >>>>> Nak. >>>>> >>>>> We already have a node in the dts for the PCI controller. Lets update >>>>> the platform code to add the pci controller to the >>>>> of_platform_bus_probe >>>>> list. >>>> >>>> I've had that idea. However it's really look strange to me to call >>>> of_platform_bus_probe() on the bus node, for which we (IMO) explicitly >>>> won't like for >>>> child devices (PCI devices) to be added to of_platform bus. Would it >>>> be suitable to just call of_platform_device_create for it (Or do i >>>> miss someth<ing)? >>> >>> Try the attached patch (lightly tested). If it works for you then >>> I'll post it for wider review. >> >> Yes, this patch worked for me. However it looks a bit like a hack for me. > > I'll probably refine it a bit before merging, but I don't think it is > a hack. It reflects the behaviour that makes sense when registering > devices hanging off the root node. If a device node is a child of the > root, then we know it isn't hanging off an i2c or pci bus, or anything > else. It is essentially a system device. > > The troublesome bit is that the root node also has memory, cpus, > chosen and aliases nodes which are not devices. In the vast majority > of cases, we want all the device nodes that are children of the root > to be registered, but we don't want to register the special nodes. > Checking for the presence of a compatible property is a pretty good > test for determining whether or not a node actually represents a > device, especially because all users of of_platform_bus_probe() seem > to be FDT users where we've been very strict about enforcing that > drivers must use the compatible property for matching to device nodes. Now it's clear to me, thanks for the explanation. BTW: On 2.6.35-rc6 I had to make 'compat' and 'match' variables const.
From d84af195dbcd99ce172bf639538231141176d402 Mon Sep 17 00:00:00 2001 From: Grant Likely <grant.likely@secretlab.ca> Date: Thu, 22 Jul 2010 13:01:11 -0600 Subject: [PATCH] of/device: Register children with a compatible value in of_platform_bus_probe() Currently, of_platform_bus_probe() completely skips nodes which do not explicitly match the 'matches' table passed in. Or, if the root node matches, then it registers all the children unconditionally. However, there are situations, such as registering devices from the root node, when it is desirable to register child nodes, but only if they actually represent devices. For example, the root node may contain both a local bus and a PCI device, but it also contains the chosen, aliases and cpus nodes which don't represent real devices. This patch changes of_platform_bus_probe() to register all nodes at the top level if they either match the matches table (the current behaviour), or if they have a 'compatible' value (indicating it represents a device). --- drivers/of/platform.c | 31 ++++++++++++++++++++++++++----- 1 files changed, 26 insertions(+), 5 deletions(-) diff --git a/drivers/of/platform.c b/drivers/of/platform.c index f3f1ec8..2ead562 100644 --- a/drivers/of/platform.c +++ b/drivers/of/platform.c @@ -709,16 +709,37 @@ int of_platform_bus_probe(struct device_node *root, rc = of_platform_bus_create(root, matches, &dev->dev); goto bail; } + + /* Register each child node if either: + * a) it has a 'compatible' value indicating they are a device, or + * b) it is specified by the 'matches' table (by name or device_type) + * If a node is specified in the matches table, then all its children + * also get registered. + */ for_each_child_of_node(root, child) { - if (!of_match_node(matches, child)) + void *compat = of_get_property(child, "compatible", NULL); + struct of_device_id *match = of_match_node(matches, child); + + /* Skip if node neither matches nor has a compatible property */ + if (!compat && !match) continue; - pr_debug(" match: %s\n", child->full_name); + pr_debug(" register device: %s\n", child->full_name); + + /* Passed the first test, register node as a platform device */ dev = of_platform_device_create(child, NULL, parent); - if (dev == NULL) + if (!dev) { rc = -ENOMEM; - else - rc = of_platform_bus_create(child, matches, &dev->dev); + of_node_put(child); + break; + } + + /* Only register child nodes if specified by matches table */ + if (!match) + continue; + + pr_debug(" register children of: %s\n", child->full_name); + rc = of_platform_bus_create(child, matches, &dev->dev); if (rc) { of_node_put(child); break; -- 1.7.0.4